Evaluation of Insect Monitoring Radar Technology for Monitoring Locust Migrations in Inland Eastern Australia

Home , Aiolopus simulatrix, Australian plague locust, Diabolocatantops, Insect migration, Mythimna convecta

Haikou Wang

A thesis submitted for the degree of Doctor of Philosophy in the Faculty of UNSW@ADFA The University of New South Wales

31 July 2007

Originality Statement

I hereby declare that this submission is my own work and to the best of my knowledge it contains no material previously published or written by another person, or substantial proportions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educational institution, except where due acknowledgment is made in the thesis. Any contribution made to the research by others, with whom I have worked at UNSW or elsewhere, is explicitly acknowledged in the thesis. I also declare that the intellectual content of this thesis is the product of my own work, except to the extent that assistance from others in the project’s design and conception or in style, presentation and linguistic expression is acknowledged.

Haikou Wang

31 July 2007

I hereby grant to The University of New South Wales or its agents the right to archive and to make available my thesis or dissertation in whole or part in the University libraries in all forms of media, now or hereafter known, subject to the provisions of the Copyright ACT 1968. I retain all proprietary rights, such as patent rights. I also retain the right to use in future works (such as articles or books) all or part of this thesis or dissertation.

I also authorise University Microfilms to use the abstract of my thesis in Dissertations Abstract International (this is applicable to doctoral theses only).

I have either used no substantial portions of copyright material in my thesis or I have obtained permission to use copyright material; where permission has not been granted I have applied/will apply for a partial restriction of the digital copy of my thesis or dissertation.

Haikou Wang

31 July 2007

iii

Authenticity Statement

I certify that the Library deposit digital copy is a direct equivalent of the final officially approved version of my thesis. No emendation of content has occurred and if there are any minor variations in formatting, they are the result of the conversion to digital format.

Haikou Wang

31 July 2007

Evaluation of Insect Monitoring Radar Technology for Monitoring Locust Migrations in Inland Eastern Australia

Haikou Wang

A thesis submitted for the degree of Doctor of Philosophy in the Faculty of UNSW@ADFA The University of New South Wales

31 July 2007

vii

0 Abstract

To evaluate the utility of insect monitoring radar (IMR) technology for long-term monitoring of insect migration, a mini-network of two IMR units in Bourke, NSW, and Thargomindah, Qld, and a base-station server in Canberra, ACT, was set up in eastern Australia. The IMR operated automatically every night under the control of a personal computer that also conducted data acquisition and processing. Digitisation of radar signals, their analysis (delimitation of echoes from background noise and adjoining echoes, followed by extraction of estimates for each target’s speed, displacement direction, body alignment, radar cross-section, and wingbeat frequency and modulation pattern), and generation of observation summaries were implemented as a fully automated procedure. Wingbeat frequency was found to be retrievable from the IMR’s rotary-beam signals, and this allowed each individual target to be characterised by its wingbeat as well as its size and shape. By drawing on ancillary information from the Australian Plague Locust Commission’s database of field survey and light trap records, the echo characters indicative of Australian plague locust, Chortoicetes terminifera (Walker), were identified. Using these, about 140 nights with detectable plague locust migrations were identified for the Bourke IMR site during 1998 – 2001 and 31 nights for Thargomindah during 1999 – 2000. Analysis of these nights confirmed that C. terminifera migrates in association with disturbed weather, especially tropical troughs, in eastern Australia. Trajectory simulation based on IMR-derived displacement directions and flight speeds allowed the identification of population movements likely to reach favourable habitats and thus to develop rapidly and possibly cause a plague. The outbreak during 1999 – 2001 most likely originated from the southeastern agricultural belt after migrations and multiplications over several generations. The IMR observations demonstrated that C. terminifera migrates over long distances with the

x Abstract wind at night and indicated that it may have an orientation behaviour that prevents it from being taken too far into the arid inland, a trait that could be highly adaptive in this environment. The two IMRs were operational for more than 85% of scheduled time during the study period and provided a wealth of information of potential value for locust management and migration research.

Acknowledgments

I am in debt to my supervisor, Dr Alistair Drake, whose inspiration and encouragement have endeavoured me to explore the unknown radar world where I finally can see the insects migrating in the darkness over hundreds of kilometres away in the remote inland eastern Australia.

I would like to express my appreciation to Prof Xianian Cheng, my MSc supervisor, who has introduced me to this fantastic field of insect migration. I also wish to register my profound gratitude to Prof Joe Riley, Dr Don Reynolds, and Mr Alan Smith. Working with them intensively near rice fields in China at night, I was impressed by their pioneer adventure, work attitude, knowledge and skills, all directed at discovering the mysterious migration mechanisms.

I especially want to thank my fellow students, Ian Harman and Tim Dean, for the friendship, joyful cooperative work experience, invaluable contributions and assistance during my study. Thanks also go to the School of Physical, Environmental and Mathematical Sciences and their support staff who made my work easier, and the Australian Plague Locust Commission where I had the access to facilities and resources vital to my research project and helpful discussions about plague locust migrations with Dr David Hunter and Mr Ted Deveson.

Generous scholarship support from the Australian Government (Australian Research Council Large Grant and Australian Government International Postgraduate Research Scholarship) and UNSW@ADFA made my study possible.

Finally, I thank my wife, Min, and my son Yineng, for their understanding, patience and encouragement. Without their support and tolerance, I would not be able to finish my study.

Table of Contents

ABSTRACT...... IX 1 INTRODUCTION ...... 1 1 INSECT MIGRATION ...... 3 1.1 Definition of Migration ...... 3 1.2 Boundaries of the Lower Atmosphere ...... 4 1.3 Advances in the Studies of Insect Migration ...... 6 1.4 Monitoring and Forecasting Migration ...... 8 2 RADAR ENTOMOLOGY ...... 14 2.1 Radars in Use for Entomological Purposes...... 15 2.2 Advances in Technological Development...... 19 2.3 Advances in Applications ...... 22 3 INSECT MIGRATIONS IN EASTERN AUSTRALIA...... 35 3.1 Environments of Eastern Australia ...... 36 3.2 Key Migratory Insects...... 38 4 STRUCTURE OF THE THESIS...... 41 2 THE AUSTRALIAN IMR MINI-NETWORK ...... 43 1 INTRODUCTION ...... 44 2 THE MINI-NETWORK STRUCTURE AND ITS COMPONENTS ...... 45 2.1 The Mini-Network Structure ...... 45 2.2 The IMR Hardware ...... 47 3 THE SOFTWARE IMPLEMENTATION ...... 51 3.1 The IMR Operation ...... 53 3.2 The AWS Operation ...... 55 3.3 Extraction of Target Parameters...... 56 3.4 Statistics of Observation Results...... 58 4 SYSTEM OPERATION AND NETWORK COMMUNICATIONS ...... 62 4.1 Scheduling of Tasks...... 62 4.2 Integration and Dissemination of the Observations...... 64 4.3 Data Archiving...... 66 4.4 Routine Inspection and Maintenance Servicing...... 67 5 SYSTEM PERFORMANCE...... 69 5.1 System Reliability...... 69 5.2 System Utilisation ...... 70 6 DISCUSSION ...... 72 3 IMPLEMENTATION OF IMR SIGNAL PROCESSING ...... 75 1 ALGORITHM OF ECHO DELIMITATION...... 76 2 PARAMETER EXTRACTION ...... 79 2.1 Parameter Extraction from a Stationary-Beam Echo ...... 81 2.2 Parameter Extraction from a Rotary-Beam Echo...... 92 3 SOME MEASURES OF IMR PERFORMANCE...... 105 3.1 Performance of Echo-Delimitation Algorithm...... 105 3.2 Measurement Accuracy...... 111 4 CONCLUSIONS...... 115 4 CHARACTERISATION OF IMR TARGETS ...... 117 1 SIZE OF AN INSECT TARGET ...... 118 1.1 Mass Estimators...... 119 1.2 Reliability of RCS Measurement by the IMR...... 121 2 SHAPE OF AN INSECT TARGET...... 126

2.1 Ratio of aa42...... 126

2.2 Ratios of aa20 and aa40...... 128

xiii xiv Table of Contents

2.3 Ratio of σ yyσ xx ...... 129 2.4 Examples of RCS Shape Factors ...... 132 3 WINGBEAT FREQUENCY OF AN INSECT TARGET...... 135 3.1 Temperature Dependence of Wingbeat Frequency...... 136 3.2 Wingbeat Modulation Pattern ...... 138 3.3 Wingbeat Frequency and Insect Size and Shape ...... 139 4 IDENTIFICATION OF CHORTOICETES TERMINIFERA FROM IMR ECHOES ...... 141 5 DISCUSSION...... 143 5 TRACING MIGRATION COURSES OF AUSTRALIAN PLAGUE LOCUST ...... 147 1 CHARACTERISATION OF PLAGUE LOCUST FLIGHT...... 148 1.1 Identification of Nocturnal Flight Activity...... 148 1.2 Characteristics of Plague Locust Migration ...... 153 1.3 Weather Systems Associated with Plague Locust Migrations ...... 157 2 ESTIMATION OF PLAGUE LOCUST MIGRATIONS ...... 159 2.1 Trajectory Analysis Method...... 159 2.2 Estimating Migration Paths of Plague Locust from IMR Observations ...... 161 3 DISCUSSION...... 176 6 ORIENTATION OF AUSTRALIAN PLAGUE LOCUST ...... 181 1 MATERIALS AND METHODS ...... 182 1.1 Migration events...... 182 1.2 Environmental data ...... 182 1.3 Statistical Analysis...... 186 2 EXAMINATION OF ORIENTATION MECHANISMS ...... 186 2.1 Evidence for Orientation in Nocturnal Flight ...... 187 2.2 True Navigation...... 189 2.3 Vector Navigation...... 191 2.4 Orientation to the Wind ...... 195 3 DISCUSSION...... 205 7 DISCUSSION AND CONCLUSIONS...... 209 1 FEASIBILITY OF REAL-TIME OBSERVATION ...... 209 2 TARGET IDENTIFICATION FROM ECHO CHARACTERS ...... 212 3 ESTIMATION OF MIGRATION PATH INCORPORATING ORIENTATION ...... 214 4 EFFECTIVENESS FOR MONITORING LOCUST MOVEMENTS...... 215 BIBLIOGRAPHY ...... 219 APPENDICES ...... 241 A. INTERACTION BETWEEN JAVA AND JAVASCRIPT ...... 241 B. MIGRATION EVENTS OF CHORTOICETES TERMINIFERA AT BOURKE...... 245 C. METAPOPULATION PERSISTENCE OF PLAGUE LOCUST DURING 1998 – 2001 IN EASTERN AUSTRALIA ...... 249 D. NOTATION AND STATISTICS OF DIRECTIONAL DATA...... 259 D.1 Direction Description...... 259 D.2 Directional Data of Insect Movement...... 262 D.3 Circular statistics ...... 263

List of Figures

Figure 2.1 Map of the IMR Mini-Network in Eastern Australia ...... 46 Figure 2.2 The IMR Unit at Bourke Airport, New South Wales...... 48 Figure 2.3 Block Diagram of the IMR at Bourke Airport, New South Wales ...... 49 Figure 2.4 The IMR Unit at Thargomindah Airport, Queensland ...... 51 Figure 2.5 Flow Chart of IMR Data Stream ...... 52 Figure 2.6 Schemes of Range Shifting (a) and Beam Positioning (b) Used by the Bourke IMR...... 53 Figure 2.7 Time-Series Profiles of Target Numbers (a) and Wingbeat Frequencies (b) at Bourke during the Night 03-04 Dec 1999 ...... 59 Figure 2.8 Time-Series Profiles of Displacement Directions (a) and Surface Weather Conditions (b) at Bourke during the Night of 03-04 Dec 1999 ...... 60 Figure 2.9 Histograms of Target Parameters at Bourke for the Night of 26-27 Feb 2000...... 61 Figure 2.10 Indices (a, c) and Profiles (b) of Target Numbers and Condensed Summary of Surface Weather (d) at Bourke for the Night of 26-27 Feb 2000 ...... 62 Figure 2.11 Front Page of the IMR Web Site of Bourke ...... 65 Figure 2.12 Front Page of the IMR Web Site for Bourke, Showing a Pop-up Explanation...... 66 Figure 2.13 Time Series of Target Numbers at Bourke for the Period June 1998 to May 2001...... 70 Figure 2.14 Time Series of Target Number at Thargomindah for the Period September 1999 to January 2001...... 71 Figure 2.15 Polar Histograms of Insect Displacement at Bourke (a) and Thargomindah (b)...... 72 Figure 3.1 Criteria of echo selection (a) and possible echoes selected (b)...... 76 Figure 3.2 Echo-Delimitation Algorithm Implemented in DE...... 78 Figure 3.3 IMR Beam Geometry: Side View (a) and Plan Views (b-d)...... 79 Figure 3.4 Flow Chart of IMR Signal Processing ...... 80 Figure 3.5 Example of Analysis of an SB Echo (1) ...... 83 Figure 3.6 Example of Analysis of an SB Echo (2) ...... 86 Figure 3.7 Screen Capture of an SB Echo Analysis by the Program DE ...... 87 Figure 3.8 Characteristics of Often-Used Windows ...... 90 Figure 3.9 Comparison of Window Functions on Periodogram Estimation...... 91 Figure 3.10 Example of a RB Signal (a) and the Signal Reconstructed from Estimated Parameters (b) .. 96 Figure 3.11 Example of Parameter Estimation from an RB echo (1) ...... 97 Figure 3.12 Example of Parameter Estimation from an RB echo (2) ...... 100 Figure 3.13 Screen Capture of an RB Echo Analysis by the Program DE...... 102 Figure 3.14 Example of Wingbeat Frequency that cannot be Extracted from an RB Echo ...... 103 Figure 3.15 Signal Patterns from Different Nights of Insect Activity and Weather...... 106 Figure 3.16 Variation of Echo Counts with Different Criteria for Echo Delimitation ...... 108 Figure 3.17 Distribution of the Distance of Closest Approach to the Zenith...... 112

xv xvi List of Figures

Figure 3.18 Histograms of RCS Measurements, Displacement and Orientation on the Night of 28 Feb 2000 at Bourke...... 113 Figure 3.19 Speed Comparison of RB and SB echoes ...... 113 Figure 3.20 Speed Comparison between Small and Large Insects against Upper Winds...... 114 Figure 3.21 Comparison of Distributions of Wingbeat Frequencies from SB and RB echoes ...... 115

Figure 4.1 Variation of Mass with the Polarisation-Averaged Radar Cross-Section a0 ...... 120 Figure 4.2 Sensitivity of the IMR at Bourke ...... 122 Figure 4.3 Mass Distribution of Targets on the Night of 11-12 Feb 1999 at Bourke...... 124 Figure 4.4 Mass Distributions of Insect Targets from Different Seasons...... 125

Figure 4.5 Distributions of aa42 for Insect Targets from Different Seasons ...... 127

Figure 4.6 Distributions of aa40 with aa20 for Insect Targets from Different Seasons...... 128

Figure 4.7 Relationship of σ xxσ yy to Length Diameter of Insect Targets...... 129

Figure 4.8 Distributions of σ yyσ xx for Insect Targets from Different Seasons...... 130 Figure 4.9 Differential RCS in Relation to Insect Mass ...... 131

Figure 4.10 Distributions of σ yyσ xx against Mass for Insect Targets from Different Seasons...... 131

Figure 4.11 Distributions of Shape Factors aa42, σ yyσ xx and of Mass for Three Nights of Heavy Migration during Summer ...... 132

Figure 4.12 Pattern Comparison of Shape Factors aa40 to aa20 and of σ yyσ xx to Mass...... 133

Figure 4.13 Distributions of Mass and of aa42 for a Mixed Population ...... 134 Figure 4.14 Shape Characters of a Mixed Population ...... 134 Figure 4.15 Comparison of Wingbeat Frequency Profiles from 11-14 Feb 1999 at Bourke...... 136 Figure 4.16 Profiles of Temperature from Cobar Upper Air Sounding at 21h for Three Nights in February 1999 ...... 137 Figure 4.17 Variation of Wingbeat Frequency with Temperature ...... 138 Figure 4.18 Distributions of Harmonic Amplitude Ratio (logarithmically transformed) for Nights in Summer and Spring ...... 139 Figure 4.19 Profiles of Mass and Wingbeat Frequency from the IMR Observations at Bourke on the Night of 15-16 Nov 1999 ...... 140 Figure 4.20 Scatter Plot of Wingbeat Frequency against Mass for the IMR Observation from Bourke on the Night of 15-16 Nov 1999...... 141 Figure 5.1 Time-series of vertical-profile Showing (a) Local Dusk Flight and (b) Emigration...... 149 Figure 5.2 Time-series of vertical-profile Showing (a) Overflight and (b) Immigration...... 150 Figure 5.3 Time-series of vertical-profile Showing (a) Emigration and Overflight and (b) Emigration and Immigration ...... 151 Figure 5.4 Time-series of vertical-profile Showing (a) Overflight and immigration and (b) Emigration, Overflight and Immigration...... 152

List of Figures xvii

Figure 5.5 Conditions of C. terminifera Take-off...... 154 Figure 5.6 Histograms for (a) Duration and (b) Layer Ceiling of C. terminifera Migration at Bourke and (c) Temperature of Ceiling Height at Cobar...... 155 Figure 5.7 Histograms of Nightly Displacements (a) and Crab Angles (b) of Migrating C. terminifera over Bourke during 1998-2001 ...... 156 Figure 5.8 C. terminifera Migrations Associated Typical Summer Weather Patterns...... 158 Figure 5.9 Habitat Map of Chortoicetes terminifera in eastern Australia...... 161 Figure 5.10 Time-series of Vertical Profile and Ground Weather on the Night of 03-04 Dec 1999 ...... 163 Figure 5.11 MSLP and Trajectories of Chortoicetes terminifera for the Night of 03-04 Dec 1999 ...... 164 Figure 5.12 Vertical Profile and Weather at Bourke on the Night of 02-03 Jan 2000 ...... 165 Figure 5.13 Trajectories of Chortoicetes terminifera for the Night of 02-03 Jan 2000...... 166 Figure 5.14 Time-series of Vertical Profile and Ground Weather at Bourke on the Night of 24-25 Feb 2000...... 167 Figure 5.15 MSLP Weather Charts at 10pm and 04am on the Night of 24-25 Feb 2000...... 168 Figure 5.16 Trajectories of Chortoicetes terminifera for the Night of 24-25 Feb 2000...... 169 Figure 5.17 Time-series of Vertical Profile (a) and Displacement Direction (b) for the Night of 15-16 Apr 2000 at Bourke...... 170 Figure 5.18 MSL Charts at 10pm and 04 am on the Night of 15-16 Apr 2000 ...... 171 Figure 5.19 Trajectories of Chortoicetes terminifera for the Night of 15-16 Apr 2000...... 171 Figure 5.20 Migration Trajectories of Chortoicetes terminifera at Bourke during 1998-1999...... 173 Figure 5.21 Migration Trajectories of Chortoicetes terminifera at Bourke during 1999-2000...... 173 Figure 5.22 Migration Trajectories of Chortoicetes terminifera at Bourke during 2000-2001...... 174 Figure 5.23 Migration Trajectories of Chortoicetes terminifera at Thargomindah during 1999-2000... 175 Figure 5.24 Trajectory Lengths of Chortoicetes terminifera Detected at Bourke during June 1998 to May 2001...... 176 Figure 6.1 Isomagnetic (a) and Isogonic (b) Maps of Southeastern Australia ...... 183 Figure 6.2 Comparison of Simulated and Radiosonde Winds at Cobar, NSW ...... 185 Figure 6.3 Orientation and Displacement Direction Profiles of Chortoicetes terminifera at Bourke on 08- 09 Mar 01...... 187 Figure 6.4 Orientation and Displacement Direction Profiles of Chortoicetes terminifera at Bourke on 28 Feb 1999 ...... 188 Figure 6.5 Orientations and Displacement Directions of Chortoicetes terminifera at Bourke on 25-26 Feb 2001...... 188 Figure 6.6 Orientations of Chortoicetes terminifera Detected at Bourke during 1998-2001 ...... 190 Figure 6.7 Displacement Directions of Chortoicetes terminifera Detected at Bourke during 1998-2001 ...... 190 Figure 6.8 Distributions of Wind Bearing during Locust Migrations at Bourke during 1998-2001...... 191 Figure 6.9 Orientations of Chortoicetes terminifera Observed at Bourke on 08-09 Mar 2001...... 192 Figure 6.10 Angles of Body Alignment of Chortoicetes terminifera to the Moon’s Azimuth ...... 193 Figure 6.11 Angles of Displacement of Chortoicetes terminifera to the Moon’s Azimuth...... 194 xviii List of Figures

Figure 6.12 Orientations and Displacement Directions of Chortoicetes terminifera at Bourke on the Night of 22-23 Dec 2000 ...... 195 Figure 6.13 Displacement Directions of Chortoicetes terminifera at Bourke in Relation to Wind Directions ...... 196 Figure 6.14 Differences of Displacement of Chortoicetes terminifera and Wind at Bourke ...... 197 Figure 6.15 Distributions of Crab Angles of Chortoicetes terminifera at Bourke...... 198 Figure 6.16 Displacement Directions, Relative to the Wind, of Chortoicetes terminifera, in Relation to Wind Speed ...... 199 Figure 6.17 Differences of Simulated and Estimated Wind at Bourke ...... 200 Figure 6.18 Angle Distributions of Locust Orientation Relative to Wind Bearing...... 201 Figure 6.19 Orientations of Chortoicetes terminifera Relative to Wind Speeds...... 202 Figure 6.20 Orientations of Chortoicetes terminifera Relative to the Wind Direction Plotted against Wind Direction...... 203 Figure 6.21 Crab Angles of Chortoicetes terminifera Plotted against Displacement Directions ...... 204 Figure 6.22 Polar Plot of Crab Angles in Relation to Displacement Directions for Chortoicetes terminifera at Bourke...... 205 Figure 6.23 Possible Orientation Scheme of Chortoicetes terminifera...... 206 Figure A-1 Flow Chart of Web Generation...... 243 Figure A-2 Distributions of Chortoicetes terminifera in 1999 – 2000 ...... 252 Figure A-3 Rain Events in Eastern Australia in Spring – Autumn of 2000 – 2001 ...... 256 Figure A-4 Conversions between Coordinates ...... 261 Figure A-5 Major Relationship Types of Insect Displacement with Wind...... 262

List of Tables

Table 1-1 Entomological Radars ...... 16 Table 1-2 Other Radars Used for Entomological Purposes...... 18 Table 2-1 Specification of the IMR at Bourke...... 48 Table 2-2 Contents of FIT file...... 57 Table 2-3 Input and Output Files of FitDG...... 59 Table 2-4 Causes of Data Loss of the IMR at Bourke...... 70 Table 3-1 Windows and Figure of Merit...... 89 Table 3-2 RB Echo Counts with Different Criteria for Echo Delimitation ...... 109 Table 3-3 Counts of Short Echoes with Different Criteria for Echo Delimitation ...... 110 Table 4-1 Two-Sample Independent t-Test of Masses of Young Chortoicetes terminifera Adults ...... 123 Table 4-2 Masses of Young Chortoicetes terminifera Adults ...... 123 Table 6-1 Difference of Simulated and Radiosonde Winds at Cobar...... 184 Table 6-2 P-Values of Randomness Test for Angles of Locust Orientation to the Moon Azimuth ...... 193 Table 6-3 P-Values of Randomness Test for Angles of Locust Displacement to the Moon Azimuth...... 194 Table A-1 Migration Nights of Chortoicetes terminifera Detected by the IMR at Bourke...... 245

xix

1 Introduction

1 INSECT MIGRATION...... 3 1.1 DEFINITION OF MIGRATION ...... 3 1.2 BOUNDARIES OF THE LOWER ATMOSPHERE ...... 4 1.3 ADVANCES IN THE STUDIES OF INSECT MIGRATION ...... 6 1.4 MONITORING AND FORECASTING MIGRATION...... 8 1.4.1 Detecting and Measuring Migration...... 9 1.4.2 Modelling Migration...... 11 2 RADAR ENTOMOLOGY...... 14 2.1 RADARS IN USE FOR ENTOMOLOGICAL PURPOSES...... 15 2.2 ADVANCES IN TECHNOLOGICAL DEVELOPMENT ...... 19 2.2.1 Hardware development...... 19 2.2.2 Data Acquisition and Analysis ...... 20 2.3 ADVANCES IN APPLICATIONS ...... 22 2.3.1 Quantification of Migration ...... 22 2.3.2 Identification of Migrating Insects...... 23 2.3.3 Migratory Processes and Flight Behaviours...... 26 2.3.4 Weather Associated with Migration...... 34 3 INSECT MIGRATIONS IN EASTERN AUSTRALIA ...... 35 3.1 ENVIRONMENTS OF EASTERN AUSTRALIA ...... 36 3.1.1 Landforms ...... 36 3.1.2 Climate...... 36 3.2 KEY MIGRATORY INSECTS...... 38 3.2.1 Noctuids ...... 38 3.2.2 Acridoids...... 40 4 STRUCTURE OF THE THESIS ...... 41

Migration plays an important role in the persistence of numerous insect species including many that are economically important pests. Understanding of insect migration, a knowledge is of crucial importance in planning insect management and conservation tactics, has been enhanced considerably over the last several decades, with some of the most significant contributions arising from observations made with entomological radars (Schaefer 1976; Riley 1979; Reynolds 1988; Riley 1999; Zhai 1999). Radar provides a means of directly detecting insects migrating in the lower atmosphere without perturbing them. Special-purpose entomological radars are able to detect intense migrating insects to a range of tens of kilometres (Schaefer 1970), and to

1 2 Chapter 1: Introduction determine the characteristics of each flying insect, such as its altitude, displacement speed and direction, body alignment, size and wingbeat frequency, at ranges of up to three kilometres (Schaefer 1970; Riley 1974). Entomological radar has also been found useful for meteorological studies, revealing the structures of small-scale wind systems (Schaefer 1976; Pedgley et al. 1982; Dickson et al. 1986; Dickson 1990; Drake 1990a), sea breezes (Greenbank et al. 1980; Drake 1982a), nocturnal low-level jets (Drake 1985b) and synoptic-scale airflows (Schaefer 1976; Riley & Reynolds 1983; Wolf et al. 1986a; Westbrook et al. 1995a) from tracking insects flying in such systems.

Entomological radars, however, have not been widely used, despite the recognition of their significant contributions to entomology (Riley et al. 1992a). Reasons for this include the high cost of system construction and maintenance, the complexity of radar observation that requires the expertise in radar system engineering and the detailed knowledge of entomological biology and ecology, and the time- consuming task of data analysis. For many years, entomological radars were employed only in short-term intensive research campaigns on the migration observation of key insect pests (Drake 1993). More recently, the development of vertical-beam entomological radars has provided a capability for long-term monitoring of migrating insects (Riley et al. 1992a; Drake et al. 1994; Chapman et al. 2003) and some insights into the measurements on radar targets (Bent 1984; Drake 1993; Smith et al. 1993). How to automate the radar operation and data analysis economically and practicably, thus providing a reliable and efficient means for long-term migration monitoring of key insect species, is the focus of the research that forms the subject of this thesis.

In the remainder of this introduction, the research advances in insect migration are briefly reviewed. Some background is provided on entomological radars and on recent advances in their technology and application. The developments in understanding of insect migration that have resulted from radar observations are also described. The migratory insect species and associated migration systems occurring in southeastern Australia, where the observations presented in this thesis were made, are introduced. Finally, the subsequent chapters, which describe the hardware system of the mini- network and its operation, the software implementation of the signal processing, the target identification with the echo signatures, and the application of the IMR in the study of locust migrations in eastern Australia, are outlined. 1 Insect Migration 3

1 Insect Migration

1.1 Definition of Migration

Insects are the only invertebrates equipped with wings and capable of powered flapping flight, which they use to conduct essential activities such as seeking shelter, locating food, avoiding predators, finding mates and searching for oviposition places. The added mobility provided by flight increases the efficiency of their activities and has helped them survive and become abundant and diverse (Dudley 2000b, 2001; Wootton 2001).

Insect movements, mainly by flight, have long been classified as either trivial movements that are associated with appetitive activities in the same habitat, or as migratory movements that are relatively undistracted flights from one habitat to another (Danthanarayana 1986; Farrow & Daly 1987). More recently, a three-category classification has been introduced (Dingle 1996):

station-keeping or ranging, i.e. a sequence of movements proximately triggered by stimuli coming from resources necessary for growth and maintained within a home range or suitable habitats nearby;

migration, i.e. an undistracted movement usually triggered by stimuli that indicate an immediate or impending shortage of resources and/or by the insect’s own endogenous rhythms; and

accidental displacement, i.e. a movement on a medium such as the wind that has been initiated involuntarily.

The term migration has been used in a variety of ways over the last several decades, but confusion on its entomological meaning was finally resolved by Kennedy in 1985 (Brady 1997). Kennedy’s definition, as restated by Dingle (1996), is:

Migratory behavior is persistent and straightened-out movement effected by the organism's own locomotory exertions or by its active embarkation on a vehicle. It depends on some temporary inhibition of station-keeping responses, but promotes their eventual disinhibition and recurrence. 4 Chapter 1: Introduction

This behavioural definition has been widely accepted by entomologists (Drake & Gatehouse 1995; Gatehouse 1997; Dingle & Drake 2007). The term dispersal, which has sometimes been used as an alternative for migration, may be better reserved for indicating population separation (inclusive of both trivial and migratory flights). Dingle (1996) argues that migration can only be defined as a behavioural process, although it has important ecological consequences.

1.2 Boundaries of the Lower Atmosphere

Insect migration occurs at altitudes ranging from about one metre through to about two kilometres. Within this range, some zones with specific characteristics can be recognised. The lowest of these, usually extending only a few metres above ground level, is the flight boundary layer (FBL) (Taylor 1974; Drake & Farrow 1988), also referred as the biological boundary layer (Farrow 1986) and the biota boundary layer (Isard & Gage 2001). Here the wind is often lighter than the insect’s airspeed and thus the insect can control its movement through its own power. The depth of the FBL depends on both the flight ability of the insect and the physical conditions within the airflow (Isard & Gage 2001). Strong-flying macro-insects (body length > 15 mm, airspeeds of 2–5 m/s) that migrate in definite directions, such as butterflies and dragonflies (Moskowitz et al. 2001), often confine their flight to this zone during daytime flight, although the painted lady butterfly Vanessa cardui (Linnaeus) and the wandering glider dragonfly Pantala flavescens (Fabricius) are recently reported conducting nocturnal migrations with possible fixed-compass navigation at high altitudes (Feng et al. 2006; Stefanescu et al. 2007). Weak-flying micro-insects (body length ≤ 15 mm, airspeed ≤ 2 m/s) can travel only short distances with their own power within their shallow FBL. The FBL may deepen at night when the surface air is calm (Farrow 1986).

The lowest 1–2 km of the atmosphere forms the planetary boundary layer (PBL), in which direct effects of the earth’s surface on atmospheric motion (especially friction) are evident, and through which the wind speed increases progressively with height (Farrow 1986). The PBL varies in depth both diurnally and seasonally. During clear days over land, the PBL becomes convective and deep (1–2 km) as the surface is rapidly warmed by solar radiation. The vertical air circulations disrupt the horizontal 1 Insect Migration 5 airflow, and thus reduce wind speeds. By dusk when vertical air movements cease, the PBL typically becomes stable and shallow, and a strong low-level (100–400 m) wind shear soon develops (Drake & Farrow 1988). A nocturnal temperature inversion, in which temperature increases rapidly with altitude, often develops closely with the wind shear under cloudless conditions when the earth’s surface is cooling down by radiation (Farrow 1981b; Drake 1982b, 1984a; Farrow 1986; Drake & Farrow 1988; Isard & Gage 2001). A low-level jet (LLJ) may form and produce wind speeds that exceed the geostrophic speed of the large-scale airflow (Drake 1985b; Beerwinkle et al. 1994; Drake 1994; Isard & Gage 2001). Micro-insects, which ascend through the FBL and stay aloft within the PBL with assistance from convective thermals during daytime, sometimes can continue migrations at night for a long distance (Farrow & Dowse 1984; Farrow 1991), remaining aloft under their own power. For migrations initiated at dusk, the insects have to ascend through the FBL with their own power to reach the fast winds higher up (Pedgley et al. 1982; Riley et al. 1983). Insects migrating at night often ride on fast winds at or above the top of the temperature inversion (Drake & Farrow 1985), and usually cover longer distances than daytime migrants (Richter et al. 1973; Drake & Farrow 1988). However, some daytime migrants, such as monarch butterflies, Danaus plexippus (Linnaeus), may soar in convective upcurrents to reach and take advantage of fast upper winds for downwind flight (Gibo 1986).

Above the PBL is the geostrophic layer, in which the wind speed and direction are proportional to the local gradient of atmospheric pressure and frictional effects of the earth’s surface are no longer significant. The high-speed winds of the geostrophic layer offer the greatest opportunity for insects to undergo long-distance displacement. By day, the altitude of the bottom of the geostrophic layer is 1–2 km (above ground level), due to strong convective mixing in the PBL. At night the stable PBL is only a few hundred metres deep and the geostrophic layer is more easily reached (Farrow 1986; Drake & Farrow 1988). However, the temperature within the geostrophic layer is often below the flight threshold for many insect species and migrating insects actively adjust their flight height to avoid unfavourable conditions (Riley 1999). Therefore, insect migrations are not common within this layer. 6 Chapter 1: Introduction

1.3 Advances in the Studies of Insect Migration

The knowledge about insect migration has been improved markedly by multidisciplinary approaches. Physiological and aerodynamic studies show that flight is the most effective form of locomotion, though it is highly energy-consuming (metabolic rates can be 50–200 times greater than that in rest status) (Dudley 2000a; Wootton 2001). Fat, which is the most efficient storage medium, is commonly used as primary fuel during migratory flight. Proteins or carbohydrates are also used as flight fuel, or only used at the beginning of flight or during short flights (Dudley 2000a). There is a trade-off though between large fuel reserve and migration ability, because the extra weight of a large reserve reduces flight speed and consumes more energy for the same distance (Dixon et al. 1993; Lehmann 2002). In consequence, migration patterns and strategies have evolved differently among different insect species, for example, small insects are often adapted to carry out single long-distance migration on prevailing winds while many other migratory insects conduct stopover flights with frequently refuelling breaks (Dingle 1996; Dudley 2001).

Migratory syndromes (Johnson 1969; Dingle 2001) are believed to be the expression (i.e. the phenotype, like wing and phase polymorphisms) of migratory genotypes; this expression may be mediated by hormones such as juvenile hormone (JH) (Dingle & Winchell 1997) and influenced by environmental factors. JH stimulates migratory flight at intermediate titres, but induces reproductive activities and oogenesis while suppressing migration at high titres (Schneider et al. 1995; Dingle 1996). In the Migratory Locust Locusta migratoria (Linnaeus), adipokinetic hormone (AKH) and octopamine control phase polymorphism, and can induce lipid mobilisation from fat bodies to flight muscles during migration or to oocytes during oogenesis (Pener et al. 1997; Pener & Yerushalmi 1998; Canavoso et al. 2001). However, the released AKH during flights is not the factor that causes the increase of JH titre after long-duration flight (Kyung et al. 2004). On other hand, JH does not directly induce the phase change (Fairbairn 1994; Applebaum et al. 1998). Hormones are believed to play an essential role in supplying fuel during flight, though the mechanism remains unknown. Migrants can be physically different from those that undertake only trivial flights; migrating populations usually are physiologically prepared, having suitable energy reserves and suspended reproductive development – oogenesis-flight syndrome (Johnson 1969; Gunn 1 Insect Migration 7

& Gatehouse 1993). The fall migrant adults of monarch butterfly have much longer lifespan (6–8 months from August/September to next March) than the adults in other generations (< 2 months); they migrate from eastern North America to overwintering sites in south-central Mexico and their reproductive diapause persists until next spring (Herman & Tatar 2001). Polymorphism in the wings and their muscle development, and variations in flight capability and fecundity, have also been found between migratory and non-migratory populations, and many of these differences may have a genetic basis expressed via the endocrine regulation (Gatehouse 1997; Dingle 2001).

Migratory insects have been found to fly in winds blowing in preferred directions (Riley & Reynolds 1996) and to show orientation ability at night (Riley & Reynolds 1986; Riley 1989; Feng et al. 2006; Chapman et al. 2008). The orientation and navigation abilities of day-flying migrants have been reviewed though mainly in insects migrating within the FBL (Srinivasan et al. 1999; Srygley & Oliveira 2001). Evidence shows that some migrants are capable of piloting in fixed direction, orientating with sun compass, earth magnetic field, or landmark features (Srygley et al. 1996; Perez et al. 1997; Perez et al. 1999; Srygley & Oliveira 2001). Monarch butterflies, for example, migrate in autumn in a wide range of wind conditions in a preferred direction towards their perennial overwintering site. They employ varying strategies of soaring, compensating for undesired deviations in a slight crosswind or downwind flight (Gibo 1986). The common green darner, the dragonflies Anax junius (Drury), simply fly with the autumn prevailing northerly winds (Wikelski et al. 2006), while the pierid Aphrissa statira (Cramer) and nymphalid Marpesia chiron (Fabricius) butterflies can compensate off-course wind-drift or detour over open water according to the sun compass (Srygley 2003). The wandering gliders, the dragonfly Pantala flavescens, always flies towards the southwest regardless of wind directions during their autumn nocturnal migrations in north-eastern China at the altitude up to 1000 m mean sea level (Feng et al. 2006), and the silver Y Autographa gamma (Linnaeus) moths actively choose seasonally favourable winds and compensate cross-wind drifts (Chapman et al. 2008). The mechanism of orientation and navigation, based presumably on chemical, visual and physical cues, is still uncertain (Dingle 1996).

Migration, the type of insect movement that this thesis is concerned with, is now viewed as an adaptation to variations in habitat condition, and as an alternative strategy 8 Chapter 1: Introduction to diapause or dormancy (Gatehouse 1997). Instead of entering quiescence or diapause to avoid conditions that are, or are about to become, unfavourable, migratory insects move substantial distances in an attempt to find a more favourable habitat (Drake & Gatehouse 1995; Dingle 1996). Migration allows insects to avoid seasonally or ephemerally harsh environments and to exploit habitats that vary in their favourability in both space and time (Rainey 1989; Farrow & Drake 1991). However, migratory flight imposes metabolic costs for the development of flight apparatus and the energy reserve of long-duration flight, risks of increased predation and habitat-location failure, and potential reproductive cost of delayed oviposition and lower fecundity (Gunn et al. 1989; Rankin & Burchsted 1992; Zera & Denno 1997). Hence, migration can be viewed as evolved opportunism (Dingle 2001), an adaptive behaviour that exploits changing habitats and that is maintained by contemporary natural selection (Drake & Gatehouse 1995; Dingle 1996).

A wide variety of techniques have been used to improve the understanding of insect migration (Reynolds & Riley 1997). Molecular markers, e.g. protein or DNA markers, have recently been used to estimate the extent of populations of flying insects (Hagler & Jackson 2001; Loxdale & Lushai 2001). Lightweight telemetry systems have been attached to migrating insects for tracking their flight trajectories in real-time (Schregardus et al. 2006; Wikelski et al. 2006; Wikelski et al. 2007). Entomological radars, the subject of this thesis, have been especially valuable for revealing migratory behaviour at high altitudes. The new technology and sophisticated equipment have provided opportunities to examine insect migration in great detail and more extensively, and have helped to deepen the understanding of earlier findings (Dingle 1996).

1.4 Monitoring and Forecasting Migration

Migration is a critical factor in the dynamics of many insect populations especially pest insects. No rational management strategy for a particular insect pest is possible without the knowledge of whether or not it is migratory (Irwin 1999). By investigating a species’ natality, development and growth, survival and mortality, and migratory activity, a basis for forecasting changes in population level and/or distribution can be established (Dent 2000). Such an inferential system will likely incorporate monitoring within crops (by bait traps and systematic sampling) at fixed locations (e.g. with suction 1 Insect Migration 9 or light traps), and by general survey (of area-wide distribution and abundance). Precise and accurate prediction of population outbreaks is the primary objective of a forecasting system and direct observation of migration (e.g. with an insect monitoring radar) appears likely to make this task more achievable. Entomological radars (Riley et al. 1992a; Beerwinkle et al. 1993; Drake 1993) and Geographic Information Systems (GIS) (Robinson 1995; Wang et al. 1997) have recently been applied to these tasks. Some migration monitoring methods and forecasting systems are summarised briefly here.

1.4.1 Detecting and Measuring Migration A variety of direct and indirect methods have been used to establish the occurrence and extent of migration in an insect population.

Mark-capture Long-range migration of insects can be established by the methods of mark- capture or capture-mark-release-recapture (Rose et al. 1985). Marking is either by an external mark, such as dye, fluorescent dust, or clipping of a specific part of a wing, or by an internal tracer, such as a dye, radioisotope, rare element, or a protein marker; sometimes two methods are used in combination. Using this method, a great number of individuals need to be marked due to the dispersion effect of migration and the consequent very low rate of capture of marked individuals at sampling sites. Capture is conducted in the expected destination regions, usually downwind (Drake 1990b; Hagler & Jackson 2001).

Capture of naturally marked migrants The only difference between this and mark-capture is that natural marks, such as an isotope, a rare element, pollens, or parasites, are present only in a particular source region, and are used to distinguish immigrant specimens from the local population. The advantage of this method is that there is no need for an artificial marking operation and the number of naturally marked individuals may be very large. However, this method is dependent on detailed information on the distribution of the natural marks (Drake 1990b).

Recent contributions to the development of understanding of insect migration, have come from the use of molecular markers, which have been used to estimate the scale of population movements (Hagler & Jackson 2001; Loxdale & Lushai 2001). 10 Chapter 1: Introduction

Synchrony of appearance in time and space Immigration can be inferred by monitoring the occurrence of adults in an area where the species cannot overwinter, or at a time before the expected emergence date of the local population, or by determining that the number of adults is higher than that which would be produced by the local population (Cheng et al. 1979; Greenbank et al. 1980; Farrow & Daly 1987). Emigration can similarly be inferred from the disappearance of adults from the local population (Drake 1990b). Monitoring field population can be done by sampling technologies, e.g. line-transact survey and other distance sampling methods, pheromone and light trapping. Interpretation of light-trap catches, however, must be with care, as the capture is affected largely by environmental factors as well as the light-trap design (light wavelength and strength, and trap structure) and the insect’s behaviours. Furthermore, the immigrants may be attracted by the light- trap only with a period of time delay after landed, and the emigrants may not be affected by the light-trap at all, i.e. less captures may occur when the local population moves out (Cheng et al. 1979; Zhou et al. 1995). The sources and/or destination regions of a migration can therefore be inferred from changes in the distributions of adult populations and their correspondence with wind transport on passing synoptic-scale weather systems.

Aerial capture Direct capture of migrating insects has been widely used to confirm expected migrations. Methods include the use of upward pointing light traps, suction traps, light traps and nets mounted on a ship at sea or nets suspended from a kite, kytoon, aircraft or a remotely piloted model plane (Schaefer 1979; Greenbank et al. 1980; Farrow & Dowse 1984; Drake & Farrow 1985; Riley et al. 1990c; Wolf et al. 1990; Burt 1998; Shields & Testa 1999). Aerial sampling at high altitudes has been used to validate the hypothesis that a species is migratory and to establish the identity of targets detected by entomological radar (Deng 1981; Riley et al. 1987; Drake 1990b; Chapman et al. 2004).

Direct observation Visual observation has been used to study insect take-off, landing, and migratory flight, with and without the assistance of optical or opto-electronic devices such as binoculars, telescopes, and night-vision goggles (possibly with infra-red illumination); observations may be made against the moon or the twilight sky, or in the light from a 1 Insect Migration 11 spotlight (Pruess & Pruess 1971; Greenbank et al. 1980; Hunter 1981b; Riley et al. 1983; Lingren et al. 1986; Drake 1990b; Riley et al. 1990a; Wolf et al. 1990; Feng et al. 2006). However, direct observation is often limited by the target size and range, and methods using visible light may alter the flying insects’ behaviour. Technical advances in the development of miniaturised radio transmitter have made tiny tags (ca 300 mg) available for tracking individual migrating insects with a receiver on a car or an airplane (Schregardus et al. 2006; Wikelski et al. 2006; Wikelski et al. 2007). Radio tracking has extended the observation range of human eye on large insects like butterflies and dragonflies. These methods are effective for flights of migrants within the FBL, but at higher altitudes detection of small targets by these means becomes difficult.

Entomological Radar Entomological radars have significantly extended the human vision capability. With scanning radar, detection ranges are sufficient for most migrating insects, while the development of millimetre-wavelength radar has improved the detection of small insects from a few hundred to a thousand metres (Riley 1992; Yang et al. 2008). The development of vertical-beam radars has reduced the observation complexity and allowed automated operation (Riley & Reynolds 1997; Reynolds 1998). Automated radars can provide nearly real-time information about overflying insects (Riley et al. 1992a; Beerwinkle et al. 1993; Drake 1993; Smith et al. 1993; Beerwinkle et al. 1995; Smith & Riley 1996; Riley & Reynolds 1997; Riley & Smith 1999; Drake et al. 2000; Riley et al. 2000; Drake et al. 2001; Drake 2002; Drake et al. 2002a), and therefore allow long-term, routine monitoring of airborne insect faunas.

1.4.2 Modelling Migration Modelling is commonly employed to improve understanding of population dynamics. The occurrence of non-migratory insect pests can usually be predicted from their phenology, historical statistics, simple or multiple statistical regressions of biotic and abiotic factors, or accurate predictive models based on life tables. With migratory insect pests that have a high mobility to colonise or escape changing habitats on a large scale, occurrence is unlikely to be predicted accurately without considering population distributions and structures (Hamilton et al. 1994; Drake 1998). Numerical models have been developed for quantitative forecasting of population dynamics and distributions over a range of temporal and spatial scales (Rochester et al. 1996). For migratory insect 12 Chapter 1: Introduction pests, numerical forecasts need to incorporate estimates of population inflows and outflows due to adult migration (Farrow & Daly 1987; Jeger 1999), and requires considerable computational capability to process the large amount of data. Geographic information systems have been employed for integrating data management and exploration, spatial analysis and modelling (Liebhold et al. 1993; Robinson 1995; Deveson & Hunter 2002).

Trajectory analysis Since much migration, of both strong- and weak-flying insect species, is downwind, trajectory analysis of air particles has been used to estimate the flight paths, either forwards to destinations or backwards to sources (Riley & Reynolds 1979; Drake & Farrow 1983; Wolf et al. 1986b; Wolf et al. 1986a; Rosenberg & Magor 1987). Synoptic weather maps show airflows on a large scale and thus indicate possible migration pathways. Migration trajectories, therefore, can be calculated from wind fields estimated from routine meteorological observations by incorporating flight parameters for the species such as the time of take-off, flight duration, transportation height, and temperature threshold. Airborne entomological radars and constant-altitude balloons have validated the flight trajectories estimated from wind fields (Hobbs & Wolf 1989; Westbrook et al. 1995a).

A regional simulation model of medium-range movements in Australia by Helicoverpa moths within cropping areas was incorporated into HEAPS, HElicoverpa Armigera and Punctigera Simulation, based on flight parameters estimated from radar observations, field surveys and laboratory experiments (Fitt et al. 1995). The Helicoverpa moths, H. armigera (Hübner) and H. punctigera (Wallengren), are assumed to fly at the same ground speed and duration without habitat selection when the wind speed is greater than their flight threshold of 1.0 m/s. A heading angle to the wind direction, which is assumed to be inversely proportional to wind speed, can be set for simulation of crosswind movements. The model considers the change of wind direction and speed during each moth’s flight. However, the model does not take into account the vectorial addition of wind speed and the moth’s air speed, nor the dispersion of the population during movement (Dillon et al. 1996). A spatial simulation model of the long-range migration of Helicoverpa moths from the inland to the cropping areas further east has also been developed (Rochester et al. 1996). It uses a three-dimensional 1 Insect Migration 13 trajectory analysis to simulate the moth’s movements according to their physical condition and the environmental situation, such as the moth’s flight ability that is affected by its host types and stages, and the wind speed that affects the travel distance. Use of randomly generated flight parameters may not be the best description of the obligate migration of H. punctigera. In addition, the moths may redistribute themselves into suitable hosts when the long-distance migration terminates in unsuitable habitats. The HEAPS, the long-distance migration model, and a model of distribution and abundance based on climate forecasts and updated with observations of rainfall and of vegetation (from satellite imagery), form a complete forecasting system for Helicoverpa spp. in Australia.

GIS and Decision Support Systems With a powerful geographical information system, all available information on population development and distribution can be integrated. In Desert Locusts Schistocerca gregaria (Forskål), swarms arise from either gregarious populations or concentrated solitarious populations following above-average rains, and both solitarious and gregarious populations migrate downwind (Magor 1995). A remote-sensing system, ARTEMIS (Africa Real Time Environmental Monitoring Information System), is used to assess rainfall events and to monitor vegetation conditions (Cherlet et al. 1991). Subsequently, a GIS-based system, SWARMS (Schistocerca WARning Management System), has been built for monitoring and forecasting the Desert Locust. It uses databases of digitised historical records of locust infestations and climate records for more than 60 years, real-time satellite images, weather data, and locust reports (Healey et al. 1996; Cressman 1997; Magor & Pender 1997; Voss & Dreiser 1997).

A forecasting system for outbreaks of the African armyworm Spodoptera exigua (Hübner) in East Africa has been developed based on statistical studies of the relationship of historical outbreaks to weather, and on trajectory analysis of secondary outbreaks from the primary sources. The expert system, based on estimates of rainfall from real-time satellite images of cloud-top temperatures and field observations of population distributions, is built for the ‘strategic control’ of primary outbreaks, with the aim of preventing secondary outbreaks (Odiyo 1990). Analysis of rainstorm distributions from satellite imagery improves the forecasts and reduces the survey areas required for confirming where moths are concentrated by storm airflows (Tucker 1997). 14 Chapter 1: Introduction

Forecasts of Australian plague locust Chortoicetes terminifera (Walker) outbreak are made by the Australian Plague Locust Commission (APLC) using a decision-support system (DSS) based on the GIS platforms of Arc/Info and ArcView (Environmental Systems Research Institute, Redlands, CA, USA, http://www.esri.com) (Deveson & Hunter 2002). The system incorporates models to simulate population development and dynamics (including egg quiescence and diapause), and uses wind trajectory models to estimate possible migration routes. Data inputs include meteorological data (regional observations and forecasts) downloaded automatically from the Australian Bureau of Meteorology using File Transfer Protocol (FTP) via the Internet, satellite images of vegetation conditions derived from NOAA AVHRR images [normalised difference vegetation index (NDVI) images] also via FTP, field survey data on locust occurrence and habitat conditions sent from field palmtop computers via high- frequency radio modems, locust light-trap catches from the APLC charted operators, control records with GPS coordinates from the log files of spray planes, and witness reports from the public. The system is used to direct field surveys, to forecast outbreaks, and to assess control effects in eastern Australia (Hamilton 1998; Deveson 2001). An emergency response system for monitoring outbreaks of the plague locust in Western Australia has also been developed on a GIS platform, GeoMedia Web Map (Intergraph Corporation, Huntsville, AL, USA, http://www.intergraph.com), at the Department of Agriculture of Western Australia. It is a distributed system based on the Microsoft (MS) Windows platform. Data can be entered in regional offices using MS Access and query results can be displayed using MS Internet Explorer (Beeston et al. 2001).

2 Radar Entomology

‘Radar entomology is the technique and science of using radar to study insect movement’ (Drake 2001). It began in 1968 when G W Schaefer built the first radar specifically to observe Desert Locusts and operated it just to the south of the Sahara in West Africa (Roffey 1969; Schaefer 1969; Reynolds 1988). This research was subsequently taken up by J R Riley, based in the United Kingdom but working mainly in Africa and Asia (Riley 1974, 1999). In the following years radar entomology groups were also established in Australia (Reid et al. 1979; Drake 1981a), the United States 2 Radar Entomology 15

(Richter et al. 1973; Wolf 1979a, 1979b), and China (Chen et al. 1985). Several types of entomological radar have been developed, and used to study a variety of migratory insects. Radar entomology has made much progress in its technical development and contributed greatly to the understanding of insect migration and flight behaviour over the past three decades (Riley 1999; Zhai 1999; Chapman et al. 2003).

2.1 Radars in Use for Entomological Purposes

Strictly speaking, only radars that are specifically designed for the sole purpose of detecting flying insects are termed entomological radars. However, insects are sometimes observed, and occasionally studied, with radars developed and used for other purposes. In this section the radar types used for entomological observations are summarised.

Table 1-1 lists all types of entomological radar that have been specifically built and used solely for studies of insect flight. The great majority are 3.2-cm wavelength (X-band) units, due to the fact that reliable transceivers are commercially available at this frequency at reasonable cost. X-band radars have been widely used as their wavelength is suitable for observing large migratory species of insects. A notable exception is Ka-band scanning entomological radars (wavelength 8.8 mm), which were designed for monitoring small insects like aphids and planthoppers and proved to be very successful at this task. Currently, there are two Ka-band units only, one in UK (Riley 1992) and the other in China (Yang et al. 2008). All entomological radars employ pulsed transmissions, except the single bistatic radar which was designed for very short-range observations.

Other radars that have been temporarily used to study insect flight are listed in Table 1-2. These units, mainly weather and military radars, employ a variety of wavelengths. Although these units were not designed for observing insects, their high power, locations, and sometimes also their simultaneous operation can make them very powerful for monitoring large-scale migrations. Locusts and aphids have been observed successfully with weather radars.

16 Table 1-1 Entomological Radars Type Transceiver Antenna Data-acquisition Study object parameters (CPR diameter, feed, facility (insect, research location and year) (WL;PD;PRF;PPP) BW, polarisation) (view & recording) 3.2 cm; 0.1, 0.25, 0.9 m, dipole, 2.5°, PPI, 16 mm cine Desert Locust, Niger, 68; grasshoppers, Sudan, 71; Australian plague locust, 1.0 µs; 2.0, 1.0 HP, 23 RPM, –10 ~ camera; A-scope, Australia, 71 (Roffey 1969; Schaefer 1969, 1970; Roffey 1972; Schaefer kHz; 25 kW 90°Z tape recorder 1976). 3.2 cm; 0.05, 0.1, 0.9, 1.2 m , dipole, PPI, 16 mm cine Desert Locust, Saudi Arabia, 72, Mali, 73-75; grasshoppers, Mali, 73-75, 78; 0.25, 1.0 µs; 2.0, 2.5, 1.8°, HP, 22 camera, A-scope, African armyworm, Kenya, 79,82; rice leaf roller, China, 88-91; planthoppers, 1.0 kHz; 20 kW RPM tape recorder Philippines, 83-84, China, 88-91 (Riley 1974; Reynolds & Riley 1979; Riley & Reynolds 1979; Riley et al. 1981; Riley & Reynolds 1983; Riley et al. 1983; Rose et al. 1985; Riley et al. 1987; Reynolds & Riley 1988; Riley & Reynolds 1990; Riley et al. 1990c; Riley et al. 1991; Riley 1992; Riley et al. Chapter 1:Introduction 1994; Riley et al. 1995a). 3.2 cm; 0.05, 0.25, 1.5, 1.8 m, double- PPI, 16 mm cine Australia, Australian plague locust, 74, 79 (Reid et al. 1979; Drake 1981a, 1.0 µs; 3.4-0.85 dipole, 1.4°, HP, 20 camera, A-scope, 1982b, 1983; Drake & Farrow 1983); Insect migration & weather patterns, 73- kHz; 25 kW RPM range gates, tape 91 (Drake et al. 1981; Drake 1982a, 1984a, 1984b, 1985a, 1985b; Drake & X-band recorder Farrow 1988, 1989; Drake 1990a, 1991). 3.2 cm; 0.1 µs; 1.5 m, 1.45°, HP PPI, 35 mm camera Spruce budworm, Canada, 73-76 (Schaefer 1976; Greenbank et al. 1980). 1760 Hz; 25 kW 3.2 cm; 0.05, 0.25, 1.2 m, 1.65°, HP, 60- PPI, 16 mm cine USA, corn earworm, 80-92; tobacco budworm, 79,82; fall armyworm, 82, 84; 1.0 µs; 3.4-0.85 20 RPM, 0 ~ 88°Z camera, 35 mm honeybee drones & workers, 85- (Wolf 1979b, 1979a; Lingren & Wolf 1982; kHz; 25 kW camera, A-scope Wolf & Pair 1982; Loper & Wolf 1986; Wolf et al. 1986a; Loper et al. 1987; Pair et al. 1987; Sparks et al. 1987; Beerwinkle et al. 1988; Loper et al. 1988, 1989, 1990, 1993; Wolf et al. 1993b; Westbrook et al. 1995a). 3.2 cm; 0.08 µs; 1.5 m, 1.5°, HP, 8 PPI, 35 mm camera China, oriental armyworm, meadow moth, 84-86 (Chen et al. 1985; Chen et 3.0 kHz; 10 kW RPM,-1~60°Z,4CPM al. 1988; Chen et al. 1989; Chen et al. 1992). 3.2 cm; 0.08 µs; 1.5 m, 1.5°, HP, 6-7 Digital PPI, VGA China, cotton bollworm, beet armyworm, clover cutworm, carabid beetles, 2.0 kHz; 10 kW RPM frame grabber, PC dragon flies, 00- (Cheng et al. 2002; Feng et al. 2003; Feng et al. 2006; Feng Ground-based pencil-beam conical-scan radar Ground-based pencil-beam et al. 2007; Zhang et al. 2007b; Zhang et al. 2007a; Zhang et al. 2008). 8.8 mm; 0.1 µs; 2.0 1.5 m, horn, 0.4°, PPI, 16 mm cine Rice leaf roller, China, 88-91; planthoppers, Philippines, 83-84; China, 88-91, Ka-band kHz; 45 kW HP, 20 RPM camera 07 (Riley et al. 1987; Riley et al. 1990c; Riley et al. 1991; Riley 1992; Cheng et al. 1994; Riley et al. 1994; Riley et al. 1995a; Yang et al. 2008) X-band 3.2, 1.6 cm; 0.1 µs; 1.5, 0.7 m, 20 RPM PPI UK, bumblebee & honeybee, 95, 96; cutworm, 96 (Riley et al. 1996; Osborne 2 kHz; 25 kW et al. 1997; Riley et al. 1998; Osborne et al. 1999; Riley et al. 1999; Williams harmonic et al. 1999; Capaldi et al. 2000; Svensson et al. 2001)

Shipborne 3.2 cm, 25 kW 1.2 m A-scope Fall armyworm, USA, 82, 83 (Wolf et al. 1986b). 3.2 cm; 0.1 µs; 0.6, 0.9 m, dipole, 32 range gates, A- Spruce budworm, Canada, 75-76 (Schaefer 1979; Greenbank et al. 1980). Airborne 2048, 1760 Hz; 25 3.5°, 2.4°, RLP, 8 Hz scope, strip-chart, nadir- kW tape recorder pointing 3.2 cm; 0.05, 0.10 0.6 m, dipole, 2.5°, 48 range gates, A- USA, corn earworm, 87-89, fall armyworm, 89 (Hobbs & Wolf 1989; Wolf et rotating beam µs; 2560 Hz; 25 RLP, 10 Hz scope, strip-chart, al. 1990). kW tape recorder/PC 3.2 cm, 0.1 µs, 1.2 m, dipole, 1.8°, Range gates Grasshoppers, Mali, 75,78 (Riley 1978; Riley & Reynolds 1979, 1983). 1.25 kHz, 20 kW RLP 3.2 cm; 0.08 µs; 3 1.8 m, dipole, 1.2°, PC, raster-scan USA, corn earworm, 92, aerial insect activity, 92 (Beerwinkle et al. 1993; g beam Rotatin kHz; 10 kW RLP, 0.4 Hz cathode ray tube Beerwinkle et al. 1995; Westbrook et al. 1995a). 3.2 cm; 0.1 µs; 1.5 1.2, 1.5 m, double- PC, 15 range gates Desert Locust, Mauritania, 93, 94 (Riley & Reynolds 1997); Australian plague locust, 91 (Riley 1992; Riley et al. 1992a; Smith & Riley 1996); Cotton kHz; 25 kW dipole (offset 0.1 Entomology Radar 2 BW), 2.6°,1.4°, RLP, bollworm, India, 85-86 (Reynolds & Riley 1997); UK, diamondback moth, 00 6.25 Hz (Chapman et al. 2002b), silver Y moth, 03 (Chapman et al. 2008), carabid beetles, (Chapman et al. 2005), green lacewing, (Chapman et al. 2006); Aerial fauna & sampling, UK, 1995- (Riley et al. 2000; Chapman et al. 2004; Reynolds et al. 2005; Wood et al. 2006; Reynolds et al. 2008). 3.2 cm; 0.07 µs; 1.8 m, dipole (offset PC, 16 range gates Australia, locusts, 93-; corn earworm, USA, 96 (Drake 1993; Drake et al.

X-band vertically-pointing linear-polarised radar X-band vertically-pointing 1994; Drake et al. 1998; Drake et al. 2001; Drake 2002; Drake et al. 2002b;

Ground-based zenith-pointing Ground-based zenith-pointing 2560, 1250 Hz; 25 0.7°), 1.1°, RLP, 10- Tilted rotating beam kW 5 Hz Drake et al. 2002a; Harman & Drake 2004; Wang & Drake 2004; Dean & Drake 2005). 3.2 cm; 0.05, 0.1, Orange peel, 3.9°H × A-scope Grasshoppers, Mali, 74, 78; African armyworm, Mali, 74; Kenya, 79, 80, 82 X-band RHI radar 0.25, 1.0 µs; 2.0, 1.5°V, 60°Z, 20CPM (Riley et al. 1981; Riley & Reynolds 1983; Riley et al. 1983; Reynolds & 1.0 kHz; 20 kW 1.2m,1.8°,HP,180°Z Riley 1997). 3.2cm;1µs;100kW 2.5 m, 1.1°, VP USA, 83-85, (Mueller & Silha 1978; Mueller & Larkin 1985)(AN/GPG-1) X-band tracking 3.2 cm; 0.25, 1.0 1.8 m, VP A-scope, PC screen USA, corn earworm, insect flight and wind, 94 (Wolf & Schleider 2002). radar µs; 60 kW and disk X-band bistatic 3.2 cm, 100 mW 24 cm, dipole, 9° Cotton bollworm, India, (Riley 1988). radar transmitted power WL: wave length, PD: pulse duration, PRF: pulse repetition frequency, PPP: peak pulse power, BW: beam width of half-power, CPR: circular parabolic reflector, HP: horizontal polarisation, VP: vertical polarisation, LP: linear polarisation, DLP: dual linear polarisation, RLP: rotating linear polarisation, RPM: revolutions per minute, CPM: cycles per minute, PPI: plan position indicator, A: azimuth, Z: zenith. 17

Table 1-2 Other Radars Used for Entomological Purposes Type Transceiver parameter Aerial Data acquisition Study object (WL;PD;PRF;PPP/TP) (CPR diameter, feed, facility (insect, research location and year) BW, polarisation, a/z (view & recording) range) X-band conical-scan 3.2 cm; 1.0 µs; 300 Hz; 1.0 ° PPI, RHI Desert Locust, India, 62 (Ramana Murty et al. 1964; meteorological radar 250 kW Mazumdar et al. 1965) 10 cm; 0.05 s modulation 2.44 m, (Atlas & Harris 1970; Atlas et al. 1970; Richter et al. S-band FMCW time; 200 W 1973; Gossard & Chadwick 1979) 10.98 cm; 1.0 µs; 950 Hz; 8.5 m, 1°, DLP, 5 Hz PPI, RHI Aphids, USA, 82-88 (Mueller & Silha 1978; Mueller & S-band Doppler 600 kW Larkin 1985; Irwin & Thresh 1988) (CHILL) weather radar 10.71 cm; 1.57-4.5 µs; 8.53 m, 1°, HP, 6 Hz Corn earworm, USA, 95-96 (Westbrook & Wolf 1998) Chapter 1:Introduction 318-1403 Hz; 1000 kW (WSR-88D) (Rinehart 1997) C-band Doppler 5.33 cm; 0.6, 2.0 µs; 250 4.3 m, 0.9 °, DLP Aphids, Finland, 1988 (Nieminen et al. 2000) weather radar kW (WSR81-D) X-band 3.2 cm; 2.0 µs; 900 kW 10.4 m, 0.21° PPI, RHI Insect flight & weather structure, USA, 64-67 (Glover S-band monopulse 10.7 cm; 2.0 µs; 3000 kW 18.4 m, 0.48° & Hardy 1966; Glover et al. 1966b; Glover et al. 1966a; tracking radar Hardy et al. 1966; Hardy & Glover 1966; Hardy & Katz 1969; Hardy & Ottersten 1969) UHF 71.5 cm; 2.0 µs; 6000 kW 18.4 m, 2.9° 1.875 cm; 0.25 µs; 8.6 Parabolic cylinder, dual B-scope, strip chart, Mosquitoes, USA, 69, 70 (Frost 1971b, 1971a; Ku-band mortar kHz; 50 kW beam (2°), 1.0°H × camera Downing & Frost 1972; Frost & Robinson 1973) locating radar 0.8°V, 25°A, 11.25°Z, (AN/MPQ-4) 17 Hz 2.86 cm; 5 mW Horn antenna, VP Honeybee drones, USA, (Wolf et al. 1987) (GDM2) 2.86 cm; 65-100 mW Horn antenna, 16°, Speaker (MR-9) Circular polarisation Digital display Doppler 2.91 cm; 45 mW Slotted array, 5.5°V × Homodyne receiver (AN/PPS-15) 10.0°H, VP Range gate 0.25 µs; 4 kHz; 1 kW Elliptical CPR, 1.0°H × Receiver-display 3.5°V, HP (a/v output) 2 Radar Entomology 19

2.2 Advances in Technological Development

The technology of entomological radar has been advanced through the development of new equipment and new data processing procedures, enabling improved application to research and operational activities. Of critical importance has been the development of new types of radar, such as vertical-beam and harmonic units: these have allowed, respectively, long term studies without the continuous need of an observer, and close observation of insect flight behaviour under natural conditions.

2.2.1 Hardware development Improved radar configurations and advanced radar transceivers have enhanced radar capacities and reliability for specific observation purposes. Conventional entomological radars were mostly modified commercial marine units, which were of low cost but highly reliable and had good resolution and a useful detection range. The original linear array antenna was usually replaced with a parabolic reflector to produce a narrow pencil-beam (Schaefer 1970; Riley 1974, 1975). A radar of this type (3.2 cm wavelength) can detect individual aphids, budworm moths, bollworm moths, grasshoppers and locusts at ranges of 0.25, 0.75, 2.1, 2.1 and 3.1 km respectively (Schaefer 1970; Riley 1974; Schaefer et al. 1979; Greenbank et al. 1980; Reynolds & Riley 1988; Wolf et al. 1990), and insect concentrations at 70 km (Schaefer 1976). An X-band radar with a vertically pointing beam from a stationary antenna installed on a ship was used to measure aerial density of overflying insects by counting echoes displayed on an oscilloscope (A-scope) (Wolf et al. 1986b). Using an electronic range gating and processing system, a downward-looking airborne X-band radar was able to continuously record spruce budworm Choristoneura fumiferana (Clemens) moths at intervals of 15 m over 48 range gates (Schaefer 1979; Greenbank et al. 1980). To enhance the detection capability for small insects, Ka-band radars (8.8 mm wavelength) have been developed (Riley 1992; Yang et al. 2008). This wavelength radar can detect planthopper individuals (ca. 2 mg) at a range of up to 1000 m compared to the X-band radar range for this species of 300 m (Riley et al. 1987; Riley 1992). Attaching a diode reflector that doubles the specific transmitted frequency to an insect target and using separate transmitter and receiver antennae, produces a harmonic radar, which can monitor the tagged insect to a range of 700 m at low altitudes for flight behaviour

19 20 Chapter 1: Introduction studies (Riley et al. 1996; Osborne et al. 1997). Adding bistatic and crossed-beam (i.e. the transmitter and the receiver are separated slightly and produce a small angle between transmitted and received signals) features to an X-band radar avoiding ground clutter and allowed detection of insects flying above plant canopies at heights of less than 20 m (Riley 1988).

Because of its strong potential for application, vertical pointing radar has seen continuous technological development. With a rotating plane-polarised pencil-beam, headings (with 180° ambiguity), displacement speeds, and the wingbeat frequencies of targets can be estimated from echo time series recorded, often from multiple altitudes simultaneously (Riley 1978; Riley & Reynolds 1979; Schaefer 1979; Greenbank et al. 1980; Riley & Reynolds 1983; Hobbs & Wolf 1989; Wolf et al. 1990; Hobbs 1991; Beerwinkle et al. 1993; Hobbs & Wolf 1996). The size and shape parameters of individual flying insects could not be estimated until the development of the modern ‘ZLC configuration’ (zenith-pointing linearly-polarised conically-scanning beam) insect monitoring radar (IMR) (Riley et al. 1992a; Drake 1993). A nutation of the rotating plane-polarised vertical beam (slightly offset to the zenith axis) distinguishes these units from other entomological radars and provides additional information that allows target size and shape and the direction of movement to be determined (Bent 1984; Bent et al. 1987; Riley et al. 1992a; Drake 1993; Smith et al. 1993; Smith & Riley 1996).

A dual-dish vertical-looking X-band Doppler entomological radar has been developed at the Nanjing Agricultural University, China, and in trial on the observation of rice migratory insect pests (BP Zhai, 2007, personal communication). This system with the Doppler capability could measure the velocity component of target in the direction of the beam. Thus it could tell whether the target is ascending, traversing horizontally, or descending. With separated transmitter and receiver antennae, it can detect flying insects at very low altitude. Therefore, it eliminates the blind range in normal entomological radars.

2.2.2 Data Acquisition and Analysis Analysis of the recorded data from a scanning entomological radar, such as time-lapse cine-camera photographs of PPI or of A-scope display, or magnetic tape-recordings of echo signatures, has been a time-consuming task (Schaefer 1976; Riley et al. 1992a; Drake 1993). With traditional scanning radar operating in the surveillance mode, the 2 Radar Entomology 21 flight altitudes, track directions and ground speeds of individual targets were measured either in real time on the radar display (plan-position-indicator, PPI) or subsequently on single-frame or multiple-exposure photographs of this display. The air speed of target can be determined if a measurement of the wind (e.g. from a radiosonde ascent) is also available (Schaefer 1976; Drake 1981a). Aerial density was measured directly from counts of individual echoes, while collective orientation was recognised from the dumb- bell pattern of echoes around the PPI (Schaefer 1976; Riley 1978; Schaefer 1979; Drake 1981a, 1981b).

Improvements in radar technology and especially the incorporation of computational signal processing and digital technology have led to the acquisition and analysis of entomological radar data advancing from manual counts and calculations to fully automatic procedures. With a sequence of elevation angle changes in association with detection range adjustment at specific time intervals, a scanning radar can be used to measure the profile of aerial density efficiently (Riley 1978; Drake 1981a, 1981b). This procedure minimises the overlaps of scanning heights. Cheng et al (2002) implemented an automatic procedure for counting targets and calculating target parameters (i.e. height, direction, speed, and collective orientation) from captured images of the digital PPI of an X-band scanning radar (Zhang et al. 2006). This development has significantly reduced the workload involved in the analysis of scanning radar observations, and has already been applied in migration studies of some important agricultural pests (Feng et al. 2003; Feng et al. 2006; Feng et al. 2007). By pointing the continuously-rotated plane of a polarisation of circularly symmetric beam vertically upwards, the elongated body shape of a target can be identified in addition to its displacement direction, orientation (with 180° ambiguity), speed, and wingbeat frequency from the intensity changes of its echo, which was recorded and played back through a spectrum analyser, or digitised and then analysed spectrally on a computer (Riley 1974; Schaefer 1976; Riley 1978; Riley & Reynolds 1979). All flight parameters (speed, displacement and orientation directions) and characteristic parameters (size and shape coefficients) of individual targets are retrievable from the digitised signals recorded with the ZLC-configuration IMRs (Riley et al. 1992a; Smith et al. 1993; Smith & Riley 1996; Harman & Drake 2004). This has made it possible to monitor migrating insects automatically with an IMR over an extended period, which was one of this thesis’ aims.

22 Chapter 1: Introduction

2.3 Advances in Applications

Entomological radars have overcome the technical difficulties of observing flying insects at a distance without disturbing their behaviours (Riley 1974) and this has led to an improved understanding of insect migration and flight behaviour. The potential of entomological radar has been widely recognised and they have been applied to the study of several aspects of insect movement.

2.3.1 Quantification of Migration Early in the development of scanning entomological radars, quantitative measures of insect migration were soon developed. The aerial density of Desert Locusts was measured from counts of the echoes over the scanned volume (Roffey 1969; Riley 1975; Schaefer 1976). For estimating the target volume density from both isolated and diffused echoes, quantitative observation and analysis procedures were developed for manually operated scanning radars (Drake 1981b, 1981a), though the methods were based on the assumption that the majority of flying insects were of the same species and had comparable radar cross-sections. A computational procedure for the data processing was implemented to estimate aerial densities, speeds, directions, fluxes, and migration rates (i.e. the rate of migrants crossing a unit line), from counts of echoes in defined regions of the PPI display. Using these methods a series of nocturnal migrations of Australian plague locusts was quantified (Drake & Farrow 1983). These quantitative measures were estimated for migrating devil’s grasshoppers Diabolocatantops axillaris (Thunerg) using a similar procedure (Riley et al. 1987). A PC-based X-band scanning radar was used to study migrations of cotton bollworm Helicoverpa armigera moth, beet armyworm Spodoptera exigua moth, clover cutworm Scotogramma trifolii (Rottemberg) moth, carabid beetle Pseudoophonus grieseus (Panzer) and wandering glider Pantala flavescens dragonfly; target numbers, speeds, and displacement directions are calculated semi-automatically (Feng et al. 2001; Cheng et al. 2002; Feng et al. 2003; Feng et al. 2006; Zhang et al. 2006; Zhang et al. 2007b; Zhang et al. 2008). Vertical-pointing units (upward- or downward-looking) were used to study vertical profiles in addition to flight trajectories (Riley 1978; Hobbs & Wolf 1989; Wolf et al. 1990). Insect monitoring radars have demonstrated the capability to provide more quantitative information on flying insects automatically (Bent 1984; Riley et al. 1992a; Beerwinkle et al. 1993; Drake 1993). Recent advances include quantitative studies on 2 Radar Entomology 23

Australian plague locust C. terminifera (Drake et al. 2001; Drake 2002; Drake et al. 2002a), diamondback moth Plutella xylostella (Linnaeus), carabid beetle Notiophilus biguttatus (Fabricius), green lacewing Chrysoperla carnea (Stephens), and migration layer formations (Chapman et al. 2002b; Chapman et al. 2005; Reynolds et al. 2005; Chapman et al. 2006; Wood et al. 2006; Reynolds et al. 2008).

2.3.2 Identification of Migrating Insects Target identification has long been an unsolved problem in radar entomology. Sampling with an aerial net carried aloft by an aeroplane, a model plane, a kytoon, or a kite gave direct evidence of the identity of an airborne population (Schaefer 1976; Greenbank et al. 1980; Drake & Farrow 1983; Farrow & Dowse 1984; Drake & Farrow 1985; Riley et al. 1987; Riley et al. 1991; Chapman et al. 2002b; Chapman et al. 2004; Chapman et al. 2005; Chapman et al. 2006). Spotlight is also a direct aid in identifying large insects flying at nighttime (Feng et al. 2003; Feng et al. 2006; Feng et al. 2007; Zhang et al. 2008). When direct sampling of aerial targets was not practicable, identification of flying targets detected by entomological radar had to rely on ecological data and visual observation of take-off (Drake & Farrow 1985; Hobbs & Wolf 1989; Riley et al. 1990c). Ancillary information, such as light-trap and pheromone-trap catches (Riley & Reynolds 1983; Riley et al. 1983; Feng et al. 2003; Feng et al. 2006; Feng et al. 2007; Zhang et al. 2007b; Zhang et al. 2008) and field surveys and local reports (Drake et al. 2001) can aid target identification inferentially but some uncertainty remains. For example, if a massive emigration is observed by radar, it is generally safe to infer that it is of the predominant species present locally at young adult stage (Riley et al. 1983). It is also possible to infer the immigration from a sudden increase in number of adults with chorionated oocytes. However, it is difficult to identify overflying targets without aerial sampling.

It has often been suggested that information in the radar echo could be used, at least to some extent, to identify targets. The RCS from an insect target is a function of the insect geometric size and the radar frequency (Blake 1986). Most insects are in the Rayleigh scattering region at X-band frequencies and fine detail of their shape therefore cannot be revealed (Blake 1986). The polarisation sensitivity provides information about the general form of a target (i.e. its shape). Additionally, wing flapping introduces a modulation to the RCS, which is detectable for most insects. The RCS modulations

24 Chapter 1: Introduction probably arise not so much from the flapping wings themselves (as they are barely able to reflect microwave energy), but from body distortions as wing muscles contract and relax. Modulations attributed to abdominal movements in association with respiration have also been claimed for some very large insects like Desert Locust (Schaefer 1969, 1976). Furthermore, second and third harmonics were sometimes present in the recorded signals, even when no fundamental frequency was detected (Riley 1974). These features could be used to supplement wingbeat frequency as an additional characteristic for identifying the targets. Therefore shape and wingbeat frequency, in addition to size which can be inferred from the target’s RCS, are potentially available to help identify the target.

The target RCS provides valuable information about target size, which is a direct indicator of target identity (Roffey 1972; Riley 1973). The equivalent water-drop, i.e. a water droplet containing the same amount of free water as a live insect, was initially assumed to produce the same size RCS as the insect (Rainey 1955), but this approach has proved inadequate (Riley 1973, 1974). The target RCS is dependent on the radar’s polarisation and the target view angle, as most insects are approximately cylindrical (Roffey 1969; Schaefer 1969). In the Rayleigh scattering region, the maximum RCS is obtained when the insect longitudinal body axis is parallel to the electric field of the radar wave (Riley 1985). On the other hand, for larger insects like the Desert Locust, which is in the Mie scattering region, the maximum RCS occurs when the radar beam’s electric vector is perpendicular to the longitudinal body axis. Nevertheless, polarisation- averaged RCS has been found to be a better index to insect size in both scattering regions (Aldhous 1989).

By combining the beam nutation (wobbling) with rotating polarisation, the ZLC beam at a slight offset angle provides the mechanism to measure the target’s body size and shape (Bent 1984; Riley et al. 1992a; Drake & Gatehouse 1995), which could provide a quantitative measure of target geometry with potential for target identification

(Dean & Drake 2005). The shape character of σ yy/σ xx was used in addition to the target mass estimated from its RCS by Chapman et al. (2002b) to identify diamond moth P. xylostella migrating over England.

Modulations of the radar signal due to wingbeating occur at well defined frequencies, often allowing discrimination of targets to species, or even sex (Schaefer 2 Radar Entomology 25

1970; Riley 1974). The RCS modulations caused by wingbeating or breathing were retrieved from echo signals recorded by parking the radar beam at a suitable angle and waiting for a target to fly through it (Schaefer 1969; Roffey 1972), or by aiming the beam at a target with optical or tracking equipment (Riley 1974). Wingbeat frequency was also measured with radars installed on aircraft and ships (Schaefer 1979; Greenbank et al. 1980; Wolf et al. 1986b).

Insects are easily distinguished from other radar targets. Insects can be separated from bats and birds by their smaller RCS (mostly < 3 cm2) (Riley 1985), lower air speeds (< 6–7 m/s), and continuous wingbeating mostly at frequencies greater than 14 Hz, although some very large insects have wingbeat frequencies between 8–14 Hz that overlap those of some birds (Schaefer 1976; Riley 1979; Riley & Reynolds 1979). Bats generally fly at low altitudes, though Mexican free-tailed bats, which were observed flying at heights of up to 3000 m (McCracken & Westbrook 2002), are an exception. By analysing target trajectories and signal modulations, precipitation may be readily distinguished from individual birds and insects (Riley 1974). However, radar returns from densely concentrated insects, or from insects at long range where the radar pulse return is large, are similar to those from rain. Virga (rain falling from a cloud but evaporating before reaching the ground) presents a particular problem as it cannot be verified by a surface rain gauge or an observer (Riley 1979).

It is difficult to identify migrating insects to species level from radar signatures only, as overlap and spread of the signature parameters exist between and within species (Reynolds & Riley 1988). RCS and its modulations can only be used to classify targets when differences are distinct. Wingbeat frequency has been used to separate insect targets into groups (Roffey 1969; Schaefer 1969; Riley 1974; Schaefer 1976), but wingbeat frequency (Drake et al. 2001) and respiration rate (Schaefer 1969) are both affected by temperature. The RCSs of about 24 species have been measured either in the laboratory or the field (Riley 1973; Riley 1985; Aldhous 1989; Hobbs & Aldhous 2006), and wingbeat frequencies have been measured for about 10 species (Schaefer 1976; Riley & Reynolds 1979). Identification of targets from their echo characteristics, therefore, remains relatively imprecise. When target density is high, echo signatures are contaminated by interference from nearby targets, and direct sampling would be the better solution for identifying the migrants (Riley et al. 1991).

26 Chapter 1: Introduction

2.3.3 Migratory Processes and Flight Behaviours Insect flight has been extensively studied since entomological radars became available. Migratory processes, which can be divided into the stages of take-off, horizontal transport and landing have been studied in detail with scanning radars (Reynolds & Riley 1997; Riley 1999).

Take-off and ascent Take-off for initiating a migration is generally followed by a rapid ascent to altitudes above the FBL under the insects’ own power (Drake 1994). A mass take-off (i.e. near-simultaneous take-off by a dense population) is common in migrant insects. Many species take off at dusk, while others commence flying during mid-morning as thermal convection gets underway; take-off at dawn has also occasionally been observed (Riley 1979; Drake et al. 1981; Riley et al. 1981; Farrow & Dowse 1984; Farrow & McDonald 1987; Riley et al. 1987; Drake 1994; Chapman et al. 2003). The take-off at dusk usually starts within 30 min after sunset and reaches its peak about 25 min later (Roffey 1969; Farrow & Daly 1987). Subsequent peaks of take-off during the night have occasionally been observed in noctuids and other moths but never in acridoids (Riley & Reynolds 1979, 1983). The ascent period lasts 15–60 min (Drake et al. 1981; Drake 1994), at climb rates of 0.2 m/s for brown planthopper Nilaparvata lugens (Stål) (Riley et al. 1991), 0.25 m/s for black bean aphid Aphis fabae Scopoli, 0.4–0.5 m/s for Sudan plague locust Aiolopus simulatrix (Walker), Senegalese grasshopper Oedaleus senegalensis (Krauss) and solitarious Desert Locust S. gregaria (Schaefer 1976; Riley & Reynolds 1990; Riley 1999), 0.4–0.6 m/s for spruce budworm C. fumiferana (Greenbank et al. 1980) and African armyworm S. exempta moths (Riley et al. 1983), and 1.0–1.5 m/s for corn earworm Helicoverpa zea (Boddie) and cotton bollworm H. armigera moths (Schaefer 1976; Wolf et al. 1993b).

Day-to-day variation in take-off intensity may be due to endogenous factors, patterns of emergence and completion of the teneral period, and effects of environmental factors (Riley et al. 1983; Gatehouse 1997). Warm humid conditions, light winds, and changes of wind direction associated with the development of disturbed weather (e.g. a passing front, trough, etc.) appear to stimulate insects into initiating a migration (Drake & Farrow 1989; Burt 1998; Isard & Gage 2001). In addition, migrants may be capable of choosing winds that are blowing in favourable directions (Riley & 2 Radar Entomology 27

Reynolds 1996). High or low light intensity, strong winds or calm air, low temperatures, and heavy precipitation (or associated cold-air downdrafts) may interrupt take-off (Greenbank et al. 1980; Drake et al. 1981; Farrow 1986; Riley et al. 1995b).

Cruising altitude and layering Migrating insects are often constrained in the PBL and within layers at night. Layer heights vary among insects and change during seasons and nights (Schaefer 1976).

Some diurnal migrants, such as most butterflies and dragonflies, remain within their FBL, at heights of no more than a few metres. An exception is the monarch butterfly D. plexippus, which can ascend to more than one kilometre above the ground (Gibo 1986). The vertical convective mixing by day largely prevents day-flying migrants from concentrating into layers and reduces the distance travelled. Sub- threshold temperatures limit the ability of migrants to ascend beyond the top of the PBL, where they would have the opportunity to exploit fast winds. The aerial density gradually decreases with altitude during daytime (Johnson 1957). However, high- altitude layered concentrations (up to 2900 m) have occasionally been observed with no collocated temperature inversion, and are presumed to have been above the PBL and in some cases to be the continued flight of nocturnal migrants (Drake et al. 1981; Farrow 1981b; Drake & Farrow 1985; Drake 1994; Gatehouse 1997).

Most insects flying at night reach altitudes of a few hundred to a thousand metres. Dense nocturnal migration layers often form after the peak of take-off at dusk as the temperature inversion develops (Reynolds et al. 2005; Feng et al. 2006; Wood et al. 2006); they usually exhibit a sharply defined lower boundary and an unclear upper boundary (Drake 1984a), and occasionally a well-defined migration ceiling (Riley et al. 1991). The lowest-altitude layer is typically 50–200 m thick and often at 100–300 m above the ground (Chen et al. 1989; Gatehouse 1997). Australian plague locusts C. terminifera and native budworm H. punctigera moths flew at the top of the inversion layer or in an LLJ, at heights of less than 300 m (Drake 1984a, 1985b). Senegalese grasshoppers O. senegalensis were often concentrated at or just above the temperature inversion 100-300 m above the ground, where the temperature is 5–10 °C higher than near the surface; sometime two or three layers were seen at different altitudes (Riley & Reynolds 1979). Spruce budworm C. fumiferana moths cruised in layers, typically

28 Chapter 1: Introduction below 300 m, which were particularly well defined on nights with radiation cooling when a temperature inversion formed (Greenbank et al. 1980). During southward return migration in temperate regions, brown planthoppers N. lugens migrated in the layer (100–150 m above the ground) above a temperature inversion where the temperature was about 16° C (Riley et al. 1991). Higher-altitude layers, however, have been observed up to 1900 m at night (Drake & Farrow 1985).

Nocturnal layering of dense migrants appears to be associated with atmospheric structure (Richter et al. 1973; Reynolds et al. 2008). Layer concentrations are often located just above the temperature inversion or LLJ, where the migrants can achieve a long-distance downwind migration in relatively warm air moving at near- or even super- geostrophic speeds (Schaefer 1976; Riley & Reynolds 1979; Drake 1984a, 1985b). They were usually not in the warmest part of the vertical profile of air temperature (Drake 1984a; Riley et al. 1991). Mild turbulence may disrupt layer concentrations and therefore increases the vertical spread of the migrants (Wolf et al. 1990). The passage of an atmospheric density current can change a layer permanently, while gravity waves and solitary buoyancy waves may lift or lower layer heights, but with minor effects on the migration trajectories (Drake 1984a, 1984b, 1985a). Temperature could be the main factor that limits the altitude of nocturnal migrants.

The flying altitude of migrating insects varies within and between species, seasons, and time of day. Insects may actively ascend to high altitudes where winds are faster, or descend and therefore avoid being taken higher by updraft currents and carried extreme distances. The altitude reached, therefore, is determined by the insects’ behaviour (Schaefer 1979). Migration in layers appears to be an adaptive behaviour of flying insects to vertical weather structure (Pedgley 1990; Gatehouse 1997).

Flight speed, duration and range Nocturnal migrants actively flap their wings to maintain altitude (Schaefer 1979). There is no convincing evidence from radar studies that nocturnal migrants are carried aloft solely by air flow. Even micro-insects like aphids, if not actively flying, will descend to the ground quickly, at a speed of about 1000 m in 20 min in the absence of updrafts (Gatehouse 1997). However, up to 90% of Desert Locusts in a swarm were seen gliding for periods of up to 30 seconds (Roffey 1969); whether some nocturnal fliers glide for part of their flight, as suggested by an apparent absence of wingbeat 2 Radar Entomology 29 modulations (Riley 1979), is uncertain. This may be clarified as more radar observations become available.

Insect airspeed, i.e. the movement relative to the air, is mostly less than 6-7 m/s (Schaefer 1976). It is generally difficult to measure this parameter accurately (Dudley 2000a). With entomological radar, however, both insect displacement and wind can be measured simultaneously using a pilot balloon. Hence insect airspeed can be deduced vectorially without perturbing its flight behaviour (Riley 1974; Schaefer 1976). For example, Desert Locusts S. gregaria were measured with airspeeds of 3.5–4.0 m/s in a layer (presumedly in cruising flight), while laboratory-bred gregarious specimen flew at 5.3±1.5 m/s immediately after release (Riley 1974; Schaefer 1976). The airspeed of Sudan plague locust A. simulatrix was measured as 5.5 m/s and of spruce budworm C. fumiferana moth as 2.5±1.0 m/s (Schaefer 1976; Greenbank et al. 1980). African armyworm S. exempta moths have air speeds of 2.4±0.2 m/s in winds of 6–12 m/s but of 3.0±0.5 m/s in lighter winds (Riley et al. 1983).

Insects migrate mostly at speeds greater than their airspeed, as air motions contribute a large part of their groundspeed (Atlas & Harris 1970). Above the FBL, the wind speeds are typically much greater than insect air speeds (Riley 1975; Drake et al. 1981; Farrow 1981a; Rose et al. 1985). Gregarious Desert Locust swarms travelled at 4.5 m/s (16 km/h) during the daytime (Ramana Murty et al. 1964), while solitarious individuals flew in LLJs at speeds of 4–18 m/s (15–65 km/h) at night (Roffey 1969; Schaefer 1970). Flying insects in an LLJ can reach displacement speeds of 15–20 m/s (54–72 km/h) (Drake 1985b).

Laboratory tests and observations indicate that migrant insects can fly for several hours, and even for the whole night (Gatehouse 1997). Nocturnal migration is often more intense in the first half of the night, suggesting durations are typically less than 6 hours. Solitarious Desert Locust migrated in LLJ streams for 6 hours or more, covering more than 300 km per night (Roffey 1969; Schaefer 1970). Senegalese grasshoppers O. senegalensis flew at speeds of 35–50 km/h or even 60–70 km/h, on occasions moving more than 350 km during night-time flights (Riley & Reynolds 1979). Australian plague locusts C. terminifera were regularly observed flying at 8– 10 m/s and may have travelled 150–200 km in one night (Drake & Farrow 1983). Spruce budworm C. fumiferana moths flew 180–240 km, or even 450 km on a night

30 Chapter 1: Introduction with stronger winds, over 6–8 hr in a single night (Greenbank et al. 1980). Brown planthoppers N. lugens were observed flying from dusk until dawn, moving 200– 240 km in 12 hr (300–500 km on a night with stronger winds) (Riley et al. 1991). Wandering gliders P. flavescens had possibly flown 9-10 hr for 150-400 km to reach the island, where the entomological radar was located, in Bohai Sea (Feng et al. 2006).

However, on some occasions, migrants can conduct much longer flights. Noctuid moths have been observed to continue flying after sunrise when crossing large bodies of water at night (Drake 1985b; Wolf et al. 1986b; Wolf et al. 1986a; Westbrook & Isard 1999); a group of moths took about 4 hr to cross Bass Strait (230 km) in day time, and would have covered as much as 1000 km in a 19-hr flight if they took off at dusk the previous night (Drake et al. 1981). Desert Locust swarms have been documented crossing the Atlantic Ocean, some 4500 km in about two days, in the extreme conditions associated with strong tropical cyclones (Rainey 1989). Planthoppers can fly continuously up to 30 hr, covering 1000 km in a single trip in the temperate zone (Rosenberg & Magor 1987). Micro-insects, such as aphids, may also remain above temperature inversions and be carried along both by day and night until daytime convection erodes the inversions (Westbrook & Isard 1999). Nevertheless, these are extreme examples.

Direct evidence has confirmed that nocturnal migrants often undertake long- distance migration. Marked African armyworm moths were caught 90 and 147 km downwind one night after release; ground speeds of about 4 m/s were indicated by radar observations (Rose et al. 1985). Migrating Heliothis zea (=Helicoverpa zea) moths and other moths were followed by an aircraft equipped with an airborne entomological radar for about 400 km over 7 hr 40 min during one night (Wolf et al. 1990). Tetroon tracking indicated that corn earworm moths could travel more than 1400 km in three successive nights from southeastern Texas (Westbrook et al. 1995a). Nocturnal migrants often continue their migration journey over more than one night and thus move a long distance. While the duration of migration is generally believed to be controlled by the migrating insects (Schaefer 1979), the mechanism is not yet fully understood.

Collective Orientation Mutual orientation was recorded in photographs of daytime swarms of Desert Locust more than half a century ago (Rainey 1989). Studies with entomological radars 2 Radar Entomology 31 have revealed that collective orientation is also very common in nocturnal flights of solitarious locusts and other migrants (Schaefer 1970; Roffey 1972; Schaefer 1976; Drake 1983; Riley & Reynolds 1986). Migrating insects frequently maintain mutual body alignments, mostly downwind, but sometimes crosswind or at a fixed compass direction, during nocturnal migrations (Riley 1975; Reynolds & Riley 1979; Riley et al. 1988).

Macro-insects were often observed to have a common body alignment in nocturnal migrations downwind, crosswind, or even upwind at high altitudes (Riley 1975; Drake 1983; Riley & Reynolds 1986; Riley 1989). This behaviour has been seen in both solitarious and gregarious Desert Locusts S. gregaria (Roffey 1969; Schaefer 1969, 1970; Riley 1974; Riley & Reynolds 1979), Australian plague locusts C. terminifera (Drake 1982a), Senegalese grasshoppers O. senegalensis (Riley & Reynolds 1983), African armyworm moths S. exempta (Riley et al. 1983), and oriental armyworm Mythimna separata (Walker) moths (Chen et al. 1989). The angle between the direction of body alignment and the wind generally appears to be less than 90° and is often small (Greenbank et al. 1980; Chen et al. 1989; Beerwinkle et al. 1994). In light winds, some nocturnal migrants at high altitudes have been observed flying against the wind or in a fixed compass direction (Reynolds & Riley 1979). Spruce budworm moths C. fumiferana moths maintained their orientation downwind with an angle less than 50° before dark, and oriented completely downwind in stronger winds of up to 18 m/s, but their orientations were random in foggy and near-calm conditions (Greenbank et al. 1980). They regularly showed different body alignments to the wind at different heights (Schaefer 1976).

Micro-insects, on the other hand, only occasionally show orientation patterns (Riley et al. 1994), presumably because their airspeeds are too low to influence their movement direction except in the lightest winds when they would not migrate far anyway (Farrow 1991). Small insects, e.g. brown planthopper N. lugens and rice leaf roller Cnaphalocrocis medinalis (Guenée) (Riley et al. 1991; Riley et al. 1994; Riley et al. 1995b), were usually found to have random headings during migration, but occasionally showed some degree of common orientation (Hendrie et al. 1985; Farrow 1986).

32 Chapter 1: Introduction

Knowledge of the orientation and navigation abilities of insect migrants has advanced in recent years, though mainly for day-flying species flying within their FBL (Srinivasan et al. 1999; Srygley & Oliveira 2001). No migratory insects conduct the true return migration, i.e. the same individuals make round-trip journeys. Thus it is unlikely they develop map-based navigation from experienced familiar landmarks or grid-based navigation from learned environmental physical or chemical gradients (Bingman & Cheng 2005). Although it is doubtful whether migratory insects can conduct true navigation, i.e. using geographical features or external cues to determine relative position to goal, vector navigation, i.e. using pre-determined compass bearing for a specific period of time or distance to reach a migratory goal, may exist in some migrants (Srygley et al. 1996; Srygley & Oliveira 2001). Monarch butterflies D. plexippus, which may possess a genetic vector engrained in endogenous program, migrate in a wide range of wind conditions in a preferred direction with varying strategies of soaring, compensating for undesired deviations in a crosswind, and downwind flight to complete their autumn migration to a perennial overwintering site across the 3000 km of North American land (Gibo 1986). Nocturnal migrating dragonflies P. flavescens and silver Y months A. gamma showed the ability of compensating wind drifts for the displacements to pre-determined directions (Feng et al. 2006; Chapman et al. 2008). However, diurnal migrating dragonflies A. junius and many nocturnal migrating insects seem to simply follow the preferred wind (Riley 1978; Riley & Reynolds 1979, 1983; Wikelski et al. 2006).

The mechanism of orientation and navigation, based presumably on chemical, visual, and physical cues, is still uncertain (Riley & Reynolds 1986; Riley et al. 1988; Dingle 1996). Radar observations indicate that individual nocturnal flying insects are typically 30 to 150 m distant from each other, and they therefore almost certainly do not orient themselves by visual reference to each other, but they do seem able to sense their movement against the ground to some extent (Riley & Reynolds 1986; Riley 1989). Landing response or centring response may exist in some insect activities (Srinivasan et al. 1999). Collective orientation by macro-insects may be an adaptive behaviour (Riley et al. 1983). Its importance is little understood, but it will have some effect on migration path and destination (Riley & Reynolds 1979). The extent of the orientation may affect the accuracy of migration trajectories simulated from meteorological air-particle 2 Radar Entomology 33 trajectory models (Westbrook et al. 1995b), which have often been used in migration studies (Rose et al. 1985; Beerwinkle et al. 1994; Westbrook et al. 1995a).

Concentration and Dispersion Many reported migration events have involved densely concentrated populations, as these are often detected with relative ease. Concentrations are often formed in convergent wind systems (Reynolds & Riley 1979; Drake 1982a; Pedgley et al. 1982; Riley et al. 1987; Drake & Farrow 1988; Pedgley 1990). Desert Locusts S. gregaria were found to be highly concentrated at small-scale wind gust-fronts at night (Roffey 1969; Schaefer 1969). Grasshoppers and moths were seen to have accumulated at the Intertropical Convergence Zone (ITCZ) (Schaefer 1976; Riley & Reynolds 1983). Spruce budworm C. fumiferana moths were observed concentrated at sea breeze fronts, which appeared as line echoes on the radar’s PPI display (Schaefer 1979; Greenbank et al. 1980; Dickson 1990). Line echoes caused by insects concentrated in rainstorm out-flows have frequently been observed (Schaefer 1976, 1979; Greenbank et al. 1980; Drake 1990a), e.g. from African armyworm S. exempta moths in the spreading downdraft from a rainstorm (Riley et al. 1983). Concentrations of African armyworm moths in rotors (topographic eddies) (Pedgley et al. 1982) and gravity waves have also been seen (Schaefer 1976; Riley et al. 1983; Drake 1984b, 1985a). Day-flying dragonflies, butterflies and Desert Locust swarms were seen concentrated in polygonal cells by thermal convection (Schaefer 1976).

Both migrants and passive airborne particles such as pollens and seeds mainly follow the airflows in convergence zones, but pollens and seeds do not form concentrations (Riley & Reynolds 1990). Therefore, concentration of migrants at convergence zones must result from an active behaviour or passive response of the insects to the weather system (e.g. avoiding cold air) (Drake & Farrow 1988; Pedgley 1990). Migrations that lead to concentration of a population may be an adaptive behaviour for exploiting favourable weather patterns; for example, Desert Locust and grasshoppers that concentrated in horizontal wind convergences that were associated with rain storms in the ITCZ often reached the rain-affected habitats (Riley & Reynolds 1983; Reynolds & Riley 1988; Drake & Farrow 1989). The arrival of such concentrations could cause severe outbreaks (Riley & Reynolds 1990).

34 Chapter 1: Introduction

Without convergent wind systems or self-aggregation, migrants are usually dispersed during migration (Riley et al. 1983; Riley 1990; Wolf et al. 1990). Grasshoppers in a line concentration decreased their aerial density by ~50% when passing between two radars 102 km apart downwind (Riley & Reynolds 1983). However, whether migrants preferentially migrate in convergent or divergent weather systems is yet to be studied.

Descent and Landing Landing behaviour of migrants is difficult to observe by entomological radar due to the dispersion of the population after long-distance migration and the short descent time at low altitudes. A general descent of nocturnal migrants at, or shortly after, dawn is often observed in radar studies (Wolf et al. 1986b). Heavy precipitation could wash out migrating insects, especially micro-insects (Westbrook & Isard 1999). It seems that termination of migration by precipitation is a passive process, in which the migrants are deposited by downdrafts. However, it was observed that migrating spruce budworm C. fumiferana moths fell vertically into tree canopies with wings folded, or landed during prefrontal convective storms at high density (Greenbank et al. 1980). In addition, African armyworm S. exempta moths descended immediately when entering the vicinity of rainfall where downdrafts did not exist (Riley et al. 1983). Therefore, the descent of migrants from cruising height to ground may be an active behaviour with high adaptive value; it may be a response to either endogenous factors or environmental cues (Riley et al. 1983; Drake 1994).

2.3.4 Weather Associated with Migration Though insect migration is an active process, the weather may influence it at any stage, by inhibiting or initialising take-off, dispersing or concentrating the population, and terminating or extending migratory flight (Wolf et al. 1986b; Drake & Farrow 1988). Synoptic-scale wind systems, especially depressions and anticyclones, provide steady wind transport in particular directions (Drake & Farrow 1988), while global-scale disturbances such as the El Niño-Southern Oscillation phenomena (Tapper & Hurry 1993; Zhang & Li 1999) may affect the trends of major migration patterns. The strong winds in an LLJ increase migration distances significantly (Wolf et al. 1986a). While exploring the effects of weather on insect migration, radar entomologists found that migrating insects could be used to “trace” wind currents. This facilitated improved 2 Radar Entomology 35 understanding of meso- and micro-scale weather systems that are difficult to observe with normal meteorological networks (Schaefer 1976; Drake 1982a; Pedgley et al. 1982; Sparks et al. 1985; Dickson 1990; Drake 1990a; Pedgley 1990; Westbrook & Isard 1999).

Radar studies have shown that spruce budworm C. terminifera moths migrate in both northerly and southerly winds. It was found migrating on the northerly airflows to the west of an anticyclone and often immediately ahead of a cold front or low-pressure trough. They were also found on the persistent southerly airflow in the wake of a strong depression, which had been maintained by a stationary anticyclone in the Great Australian Bight (Drake & Farrow 1983).

Insects fly in the lower atmosphere, where wind, temperature and humidity are the main abiotic factors affecting their activity and movement. Utilising the wind, insects migrate en masse over long distances, relocating to new habitats whether these are suitable or not. However, some migrant insects appear able to choose winds that transport them to favourable habitats (Riley & Reynolds 1996), a behavioural adaptation that may have resulted from the coincidence of terminating flights within suitable habitats.

An ‘airscape perspective’ has been proposed as a means for considering the interaction between migrating organisms and the atmosphere, and insect migration has been incorporated into aerobiology, ‘the study of factors and processes that influence the movement of biota in the atmosphere’ (Isard & Gage 2001), which has provided a broader perspective for reviewing this phenomena.

3 Insect Migrations in Eastern Australia

Populations of migratory insects are significantly influenced by climate and weather conditions, and they may well have adapted to the unique environment. The Migratory Locust L. migratoria, which does not often migrate in Australia, is an example and the Old World bollworm H. armigera may be another (Riley et al. 1992b; Gregg et al. 1995; Zheng & Zhang 2000). The brown planthopper N. lugens has been found to have different migration behaviours in the tropical Philippines compared with in the

36 Chapter 1: Introduction subtropical and temperate regions of China (Cheng et al. 1979; Riley et al. 1990b). Therefore migratory species may still require deep studies in Australia even when they are well known in other countries. Eastern inland Australia has been considered as an important source for several key insect migrants. The insect populations occurring there appear well adapted to this semi-arid environment. Monitoring their abundance and distribution in the arid inland source area appears critical for the practice of preventive control techniques (Hunter & Deveson 2002; Hunter 2004). Development of a component of such a monitoring system is what this thesis aims to contribute.

3.1 Environments of Eastern Australia

3.1.1 Landforms Australia is the smallest continent with least topographic variety in the world, though eastern Australia is slightly complicated in comparison with the rest of the continent. In the west is the Great Australian Peneplain, which constitutes most of Western Australia (WA), Northern Territory (NT) and north-western South Australia (SA), and consists of an extremely large level area subjected to broad uplifts (365 m above sea level, highest elevation 1500 m). In the east is the eastern cordillera, which is just inland from the east coast and extends from Queensland (Qld), New South Wales (NSW) to Victoria (Vic); it comprises elevated horsts (highest mountain 2228 m above MSL) isolated by many faults due to localised uplifts. Lower terrain, with its lowest point below sea-level at Lake Eyre (surface at approximately -15 m MSL), is situated between the western peneplain and the eastern highlands. The subterranean Artesian Basin (extending into NT, Qld, NSW, and SA, land surface below 300 m MSL) in the north, and the Murray Basin (extending into NSW, Vic and SA, land surface below 150m MSL) in the south, are bounded by the Broken Hill Upland (300 m MSL) in the west and the Cobar Peneplain (180 m MSL) in the east.

3.1.2 Climate The mainland of Australia lies between 10° and 39° S, and much of it is under the influence of the sub-tropical high pressure belt. The aridity of inland Australia is attributed largely to the subsiding air associated with this high pressure. The latitude of the sub-tropical ridge varies seasonally, centred between 30° and 35° S in winter and 3 Insect Migrations in Eastern Australia 37

35° and 40° S in summer. Therefore, under the easterly flow to the north of the subtropical ridge much of the continent is dry in winter, except along the southern and eastern coasts. In summer, the intertropical convergence zone moves southwards bringing monsoonal rain from the north-western Indian Ocean to the interior. The east coast also receives summer rain from the easterly trade wind flow off the Pacific Ocean and Tasman Sea. The rainfall pattern is thus strongly associated with season; winter rainfall predominates in the south and summer rain in the north (Tapper & Hurry 1993).

The low relief of the topography causes relatively little obstruction to the weather systems that control Australia’s climate. This reduces the aridity of the Australian desert by allowing the occasional penetration of tropical moisture deep into the continent from the warm waters of the Indian Ocean off north-western Australia. The southeastern highlands, on the other hand, interacting with the westerly winds in winter and the easterly flow in summer, produce orographic rainfall all year around and partially block moisture from penetrating further inland during summer. Therefore, the lower lands of the eastern interior are often under severe drought. Occasionally, disturbed weather from the tropics brings heavy rains into this region. Floods may ensue and the multiple channels of the streams and creeks that flow into Lake Eyre may be kilometres wide; such inundations trigger huge growth and breeding responses of vegetation. The climate of eastern Australia is subject to the influence from the El Niño- Southern Oscillation phenomenon and its associated patterns of sea-surface temperature variations over the Pacific Ocean (Tapper & Hurry 1993). El Niño events are generally associated with lower rainfall and higher temperature in winter and spring, while La Niña events often cause widespread floods (Australian Government Bureau of Meteorology 2004).

The Channel Country, which incorporates an area in the NT, a large section of southwestern Qld, the northeast corner of SA, and a small section in the far northwest of NSW, is characterised by high temperatures and generally low and highly variable rainfall. During late spring (November) and mid-autumn (April), the weather is usually controlled by a series of high pressure systems moving eastwards along latitude about 40° S. The high pressure cells are often separated by a trough or a low pressure system, which penetrates from either a large and complex zone of low pressure over tropical Australia and/or from the mid-latitude belt of low pressure which is usually well south

38 Chapter 1: Introduction of the Australian continent at about 65° S. A col is often present between high-pressure regions on the Indian and the Pacific Oceans bringing generally stable easterlies to southeastern Australia. These may be interrupted by warm northerlies and by cooler southerlies or southwesterlies during the approach and passage of a cold front. The regular development of the col system causes wind changes with temperature variations and periodic rainfall during summer, as warm moist southeasterly trade and recurved trade winds from the Pacific Ocean penetrate into the inland of southeastern Australia (Tapper & Hurry 1993).

3.2 Key Migratory Insects

The arid to semi-arid inland of eastern Australia is characterised by substantial dry periods punctuated by shorter periods of heavy rain. Heavy rains generate abundant vegetation which is an important source of migratory insects, including several key pests of agriculture including: the noctuid lepidopterans common armyworm Mythimna convecta (Walker), native budworm H. punctigera, and cotton bollworm H. armigera; and the acridid orthopteran Australian plague locust C. terminifera. Another notable noctuid migrant, the bogong moth Agrotis infusa (Boisduval) also breeds in the inland in winter, though it does not normally extend so far west (Common 1954). Two other locusts, the spur-throated locust Austracris guttulosa (Walker) in tropical grasslands of northern Australia and the Migratory Locust L. migratoria in central highlands of Queensland, are also found occasionally in the inland (Hunter 1997; Hunter et al. 1999).

3.2.1 Noctuids Helicoverpa punctigera and H. armigera are polyphagous pests of many agricultural and horticultural crops. They also colonise a wide range of wild plant hosts in both cropping and non-cropping regions. They are highly mobile and mark-capture experiments have shown that their local movements are occasionally greater than 10 km but that H. punctigera is more likely to undertake a long-distance migration (Fitt et al. 1995). H. punctigera, an obligate migrant, is widely distributed, while H. armigera, a facultative migrant and a key pest wide distributing in the Old World, is occasionally found outside its main range of the wetter eastern coastal and northern tropical regions (Zalucki et al. 1994; Fitt et al. 1995). Rainfall in the inland area in late autumn and 3 Insect Migrations in Eastern Australia 39 winter produces lush vegetation that ensures successful breeding of H. punctigera populations. The resulting populations emigrate in later winter or early spring, often into cropping areas to the southeast. These southward migrations of H. punctigera often occur in the warm airflows ahead of a cold front (Drake & Farrow 1985). They carry the moths from tropical or subtropical regions into the temperate zone. Return migrations in autumn have rarely been reported in Australia, but possibly occur (Gregg et al. 1995); for example, H. punctigera has been identified migrating in a southerly airflow in the wake of a cold front (Drake & Farrow 1985; Farrow & McDonald 1987). H. armigera, which is less commonly found in the inland, can overwinter with diapausing pupa as far south as Griffith, NSW. H. armigera frequently moves within and between cropping regions in summer and autumn when H. punctigera is hardly seen there, even in years it was abundant in spring (Gregg et al. 1995). Therefore, the inland is probably the main source of H. punctigera while unsprayed crops and wild host plants within the cropping regions may produce most of the H. armigera population (Farrow & McDonald 1987; Rochester & Zalucki 1998; Gregg et al. 2001).

Mythimna convecta is widely distributed in temperate and subtropical Australia, constrained only by extreme high or low temperatures and by habitat availability. It can cause substantial damage to cereal and grass crops and to introduced pastures in spring. Most likely it migrates from the inland into the south or the southeast of the continent in late winter to summer and returns into the inland in autumn. A severe outbreak of M. convecta occurred in the cropping areas of eastern Australia in the spring of 1983. The outbreak originated from moths that had migrated into the inland grasslands from coastal regions in airflows around tropical depressions the previous autumn, with the subsequent return movement of the much more numerous subsequent generation in the prefrontal airflows of cold fronts the next spring (McDonald et al. 1990). It remains uncertain if this species is an obligate migrant, and neither aestivation nor diapause has been observed (McDonald 1995).

Agrotis infusa does not tolerate high and low temperatures. It is an occasional pest of winter cereal crops and pastures on the Tablelands, Western Slopes and Western Plains of NSW. The Bogong moths migrate from plains hundreds of kilometres away into the mountains of southeastern Australia, where they aestivate gregariously in crevices and small caves near the peaks throughout the summer months, and return to

40 Chapter 1: Introduction their breeding grounds to mate and lay eggs later (Common 1952, 1954). Catches in association with other migrants, including in an aerial net, have confirmed that A. infusa is a long-distance migrant during spring (Drake et al. 1981; Drake & Farrow 1985).

3.2.2 Acridoids Plagues of Chortoicetes terminifera can cause severe damage to agricultural and horticultural crops and to pastures (Wright & Symmons 1987). Good winter/spring rainfall produces excellent conditions for egg development and nymphal survival in the southern inland, and plague locust populations increase when the normally dry arid interior receives summer rain (Wright 1987). Mitchell grasses Astrebla spp. and button grass Dactyloctenium radulans are the two main wild hosts of C. terminifera (Hunter 1989). A single rainfall of more than 20 mm (for barley Mitchell grass A. pectinata) or 40 mm (for curly Mitchell grass A. lappacea), at a mean monthly maximum temperature higher than 23 °C, promotes the growth of the grasses and allows them to remain green for two months: slightly longer than the time required for the locust to develop from egg to adult (Hunter 1989; Hunter & Melville 1994). Given sufficient rainfall, C. terminifera can develop four generations in the warmer northern areas of the region but only two generations are possible in the cooler southern regions (Wright & Symmons 1987). It has been demonstrated that C. terminifera often undertake long-range migrations at night, and that they also move shorter distances (tens of kilometres) as swarms by day (Clark 1969; Farrow 1977; Reid et al. 1979; Drake & Farrow 1983; Wright 1987; Drake et al. 2001). The arid interior is believed to be the major source for invasions of agricultural regions (Wright 1987). Dry conditions causing pastures to dry off may be the major population regulator. If uncontrolled, populations that build up in the inland may invade the southeastern agricultural areas and cause substantial yield losses of crops. The species does not have a permanent well defined outbreak areas or a definite migration route from the arid interior to the southeastern agricultural zone; rather, populations migrate when suitable conditions occur. Until recently, the northward migration was paid little attention though radar observations indicated that it occurred (Drake 1983; Drake & Farrow 1983), but its importance for population persistence is now recognised (Deveson et al. 2005; Deveson & Walker 2005).

Austracris guttulosa has one generation per year. Adult maturity is initiated, in late spring and summer only when day length is longer than 13 hours and there has been 3 Insect Migrations in Eastern Australia 41 sufficient rainfall and green grasses are present. Upsurges tend to be localised and confined to tropical areas. However, A. guttulosa has been recorded making a long- distance migration from western Queensland into northern New South Wales (Hunter 1997).

Locusta migratoria has 2–3 generations per year when there are normal rainfalls in spring and summer. Outbreaks are normally in the Central Highlands and adjacent areas of Queensland. Adult swarms can fly several kilometres a day, but gregarious migration at night appears to be uncommon in Australian populations (Hunter & Deveson 2002). This is a typical adaption to the Australian environment and very different from the same species in other parts of the world (Ma 1958).

4 Structure of the Thesis

In this thesis, the establishment of an operational mini-network of two insect monitoring radars in eastern inland Australia is described and the practical application to long-term monitoring and research on migration of key migratory insects is illustrated using the data collected.

Chapter 2 describes the mini-network of autonomous insect monitoring radars from which all of the primary data in the thesis has been obtained. The hardware components of this network are described as is the software developed for automated radar operation, signal digitisation and processing, and for summarising the observation results as statistical graphs for on-line publication. System performance is reviewed and examples of the observations provided.

Chapter 3 describes the software implementation of the IMR primary data analysis, as an automated routine operation. Improvements to the algorithm to ensure reliable processing, and to maximise extraction of information, are described.

Chapter 4 examines the target characters retrieved from both rotary and stationary-beam signals and explores the possibility of using them to identify the targets. Wingbeat frequency has proved insufficient to separate species under most circumstances due to the overlap between species, its temperature dependence and the

42 Chapter 1: Introduction spread among individuals. The coefficients for target size and shape, however, show distinct patterns for different target categories. With ancillary information, the echo signatures of Chortoicetes terminifera have been identified, and migration events by this species have thus been identified using these echo characteristics.

Chapter 5 reconstructs migration scenarios of Australian plague locusts in eastern Australia from the movements detected by the IMR network. The detected migrations are categorised according to the patterns of nightly time series of vertical profiles. Migration sources and destinations are estimated from retrieved flight parameters and verified by field surveys data and other ancillary information. Return migrations from agricultural regions into the arid interior in later spring and summer are identified and their role in the subsequent build-up of populations in the arid zone following summer rains considered.

Chapter 6 analyses the orientation behaviours of migrating Australian plague locusts in relation to environmental cues. Possible compass navigation and orientation to physical features are tested statistically. The power of IMR as a tool for insect migration research is illustrated.

Finally, the value of an IMR network for the practical management of key migratory insect species, and for research on their migration, is summarised in Chapter 7. Possible system improvements are discussed, and further studies on insect migration using IMRs are proposed.

2 The Australian IMR Mini-Network

1 INTRODUCTION ...... 44 2 THE MINI-NETWORK STRUCTURE AND ITS COMPONENTS...... 45 2.1 THE MINI-NETWORK STRUCTURE ...... 45 2.2 THE IMR HARDWARE...... 47 3 THE SOFTWARE IMPLEMENTATION...... 51 3.1 THE IMR OPERATION ...... 53 3.2 THE AWS OPERATION...... 55 3.3 EXTRACTION OF TARGET PARAMETERS...... 56 3.4 STATISTICS OF OBSERVATION RESULTS...... 58 4 SYSTEM OPERATION AND NETWORK COMMUNICATIONS ...... 62 4.1 SCHEDULING OF TASKS ...... 62 4.2 INTEGRATION AND DISSEMINATION OF THE OBSERVATIONS ...... 64 4.3 DATA ARCHIVING...... 66 4.4 ROUTINE INSPECTION AND MAINTENANCE SERVICING...... 67 5 SYSTEM PERFORMANCE ...... 69 5.1 SYSTEM RELIABILITY ...... 69 5.2 SYSTEM UTILISATION ...... 70 6 DISCUSSION...... 72

As noted in the introduction, insect monitoring radars (IMRs) have demonstrated incomparable advantages over scanning entomological radars for detecting overflying insects over an extended period. However, the IMR should run according to a preset schedule and the radar signal should be processed automatically. The requirements of an automatic signal-analysis procedure for an IMR is that it should delimit target echoes from background noise and adjoining echoes, and that it should extract the trajectory and identification parameters of the targets, all without user interaction. It should remain effective even if the level of background noise varies markedly, and when the migration intensity is high. It should also be capable of processing data efficiently and quickly so that results would be available within a few hours of a migration occurring. It must be very reliable and stable, and not subject to unexpected termination. It must of course be accurate, and should provide some indications of signal quality and parameter

43 44 Chapter 2: The Australian IMR Mini-Network reliability. To evaluate the IMR utility for routine monitoring of insect migrations in the remote arid area where outbreaks often originate, a mini-network of two IMR units and a server was set up in eastern Australia and a fully automatic scheme of radar operation and data processing was implemented. With modifications and improvements on both hardware design and software development, the human effort needed to maintain daily observations has been reduced to the minimum. The implementation, operation and maintenance of this system are described in this chapter. Some sample data are also presented to show the system reliability and its application potential. Some aspects of this topic, the network operation and data processing and dissemination, have already have been published (Drake et al. 2002a).

1 Introduction

Monitoring population movements of key migratory insects is essential to the practice of preventive control in modern insect pest management (Hunter & Deveson 2002). Light-traps are often used to monitor the dynamics of local insect populations and to infer migrations, but they are affected by both abiotic and biotic factors, such as wind speed and insect behaviours, and may not reflect the population changes in the field immediately, that reduce their value as population indicators (Gregg et al. 1994). Field surveys can provide direct evidence of population changes and the possible movements of migrant populations (Zalucki et al. 1994; Gregg et al. 2001), but such operations are difficult to conduct over a large area and within the limited human resources and time available. Traces of pollen and rare elements can also provide evidence of population migration (Gregg et al. 2001), but the need for knowledge of the detailed source distribution of the tracer materials makes routine use impracticable. Rainfall and vegetation monitoring with satellite imagery are useful for locating possible habitat resources, and wind and weather condition for determining if migration would have been possible and the locations of major populations, but field surveys are required to verify whether or not large outbreaks are under way (Robinson 1995; Rochester 1999). Entomological radars, on the other hand, provide a direct means of detecting population movements of migratory insects (Schaefer 1976), and therefore may have considerable potential as sources of information for practical control (Riley & Reynolds 1990). 1 Introduction 45

Only with the development of vertical-beam insect monitoring radars employing the Zenith-pointing, Linear-polarised, Conical-scan (ZLC) configuration could autonomous monitoring of migrating insects be achieved as a long-term routine operation (Riley et al. 1992a; Drake 1993). Scanning entomological radars have proven very successful at detecting and quantifying insect migrations (Schaefer 1976; Reynolds 1988; Zhai 1999). However, the high cost and complex nature of these units, and the time-consuming task of analysing data from them, have limited their applications (Drake 1993). The development of the novel IMRs (also known as Vertical-Looking Radars, Riley et al. 1992a) has overcome these limitations (Bent 1984; Riley et al. 1992a; Drake 1993; Smith et al. 1993; Smith & Riley 1996). In addition to flight height, speed, displacement direction and orientation, an IMR can provide information characterising the individual insects detected: their body size, shape, and wingbeat frequency. The small-angle conical scan of the IMR’s vertically pointing beam interrogates a much smaller volume than the scanning radar’s. The radar signals, therefore, can be digitised at a reasonably low rate and only modest data storage is required. The technical advances that provided the powerful modern PC and its huge storage media, along with PC-compatible data-acquisition boards, had made possible the fully automatic operation of the radar as well as digitisation and analysis of its received signals. The reduced complexity of the mechanical operation of an IMR made automatic observation more practicable and decreased the system running costs and maintenance requirements. Although an IMR only monitors overflying insects at one location the information it provides should still be quite broadly applicable, as studies with scanning radars have demonstrated that insect migrations are often approximately uniform over a large area (Drake 1993).

2 The Mini-Network Structure and its Components

2.1 The Mini-Network Structure

The IMR mini-network that has been set up in southeastern Australia is shown schematically in Figure 2.1. The first IMR was installed at the airport of Bourke

46 Chapter 2: The Australian IMR Mini-Network

(30.0392° S, 145.952° E, 107 m above MSL), New South Wales, in April 1998 and started to operate in May 1998. The second IMR was installed at the airport of Thargomindah (27.9864° S, 143.811° E, 132 m above MSL), Queensland, in August 1999 and became operational in September of that year. Each IMR incorporates an Automatic Weather Station (AWS). A PC that acts as the network hub is located in the project’s base laboratory on the campus of the Australian Defence Force Academy (35.3083° S, 149.194° E, 575 m above MSL) in Canberra, Australian Capital Territory. The hub PC is connected to the campus LAN and serves as a standard web server through which the observational results from the IMRs are disseminated via the World Wide Web (WWW). This PC is also used to access the IMRs, for transferring data, upgrading software, inspecting performance logs, and diagnosing faults, via a modem connection to the public telephone network. The two IMRs are about 740 km and 1200 km to the north-west of Canberra, and can be reached by driving (all on sealed roads) in about one day and one-and-a-half days respectively. The principal users of the information from this mini-network are also shown on Figure 2.1.

Figure 2.1 Map of the IMR Mini-Network in Eastern Australia Redrawn from Drake et al. (2002a) on the topographic digital elevation model for on-line GIS available from Geoscience Australia (2002). Copyright-free icons are from the Internet. 2 The Mini-Network Structure and its Components 47

The IMR locations were selected to be in the outbreak regions of several migratory insect pests, with consideration given also to the practicalities and costs of installation and maintenance (Drake et al. 2002a). This network was set up as part of an Inland Insect Migration Project to investigate the adaptive behaviours of key migratory insects in the highly variable patchy habitats arising from the region’s erratic rainfall (Drake et al. 2001). It has been suggested that the Australian plague locust Chortoicetes terminifera and the native budworm Helicoverpa punctigera often build up their populations in inland areas of eastern Australia and fly southeastwards to adjoining agricultural and grazing regions, moving several hundred kilometres in one night (Wright 1987; Gregg et al. 1995; Gregg et al. 2001). The two IMRs are about 300 km apart, a distance which could be covered by a single night’s migration so that a migrating population could possibly be detected at both sites. The Thargomindah IMR is located well inside the area where populations rapidly build up, while the Bourke unit is near its southeastern boundary. Thus the former unit monitors movements inside the region of population development, while the latter detects the migrations into adjoining cropping regions (Drake et al. 2002a). Both IMRs are situated at airports, which provide some security and access to electricity and the public telephone network. The airports are also close to towns, which provide accommodation for staff from the base laboratory on servicing trips and are also a source of local agents who do routine maintenance work and solve minor problems with help from the base laboratory.

2.2 The IMR Hardware

The two IMRs have similar configurations. The unit installed at Bourke airport (Figure 2.2) consists of a PC, an 8-channel gated peak-hold signal-conditioning and data- acquisition unit, a multi-function control unit that is instructed by the PC to operate the radar, a transceiver, and a parabolic reflector with an offset double-dipole feed that is rotated by a DC permanent-magnet motor (Figure 2.3, Table 2-1). The main electromagnetic equipment of the IMR and the control and data-acquisition PC system are housed in an air-conditioned cabin. The antenna incorporates a cylindrical metal shroud that is lined with microwave absorbent material for sidelobe shielding and enclosed by a waterproof radome. There is also an Automatic Weather Station (AWS) that is linked to the PC.

48 Chapter 2: The Australian IMR Mini-Network

Figure 2.2 The IMR Unit at Bourke Airport, New South Wales This photo shows the air-conditioned cabin that hosts the IMR equipment and its control PC; the radar antenna covered with a weatherproof radome surrounded by a security fence; and the AWS (in front). The electric power distribution box can be seen inside the fence. (December 2000)

Table 2-1 Specification of the IMR at Bourke Transmitter Peak Power 25 kW Frequency 9.4 GHz Pulse Repetition Frequency 2560 Hz Rotary, 2500 Hz Stationary Pulse length 50 ns Receiver (Logarithmic) Intermediate Frequency (IF) 60 MHz Noise figure 12 dB IF bandwidth 20 MHz Antenna (Parabolic reflector with offset double-dipole feed) Polarisation Linear, rotating Diameter 1.8 m Beamwidth (half power) 1.1° Dipole rotation rate 5.0 Hz Dipole offset (in beamwidths) ~ 0.2 Data Acquisition Gated peak-hold channels 8 Gate width 33 ns (50 m) Cut off frequency of low-pass filters 128 Hz Digitisation sampling rate 320.0 Hz Rotary, 312.5 Hz Stationary Digitisation samples per dipole rotation 64 (per channel in rotary mode) Sample resolution (A/D Converter) 12 bits

2 The Mini-Network Structure and its Components 49

Antenna Multi-Function Control Unit AWS

AWS Data Modified Acquisition Transceiver Unit

Figure 2.3 Block Diagram of the IMR at Bourke Airport, New South Wales The unit consists of an antenna, a transceiver, a multi-function control unit, an 8-channel gated peak-hold data-acquisition unit, an AWS, and a PC that is connected to the public telephone network via a modem.

A 486 DX2/66 PC controls the radar operation, data acquisition, data processing, and communications. Two data-acquisition (DAQ) boards are installed in the computer; the digital-analog (D/A) converters on them are used to control the transceiver tuning and the antenna-feed rotating and aligning, and the analog-digital (A/D) converters on them are used to digitise the radar signals and monitor the hardware performance. The multi-function control unit provides a control and monitoring interface between the radar system and the PC. The radar signals are digitised discretely from the peak-hold range-gates into the PC’s extended memory via direct memory access and then saved to the hard disk by an interrupt buffering, which avoids burdening the PC processor. The 8.6 GB capacity of the hard disk provides temporary storage for a few weeks of data. With a Compact-Disk (CD) burner, the PC archives all the raw data and the processed results to CDs once or twice each week. The PC also acts as a terminal, to control and operate the AWS through a RS232 link. Via a modem connected to the public telephone network, the PC can be accessed remotely from the base laboratory for transferring data, reprogramming and testing the radar operation.

The radar transmitting and receiving unit is from a commercial marine navigation system, a Decca type 65160 (Racal Decca Marine Ltd, Surrey, UK) operating at 9.46 GHz (X band, 32 mm wave length). It is connected by rectangular waveguide to a 1.8-m diameter, vertically pointing parabolic reflector via a double- dipole feed. The antenna produces a linearly-polarised beam of half-power full width 1.1°. The feed is slightly offset from the zenith and connected through a rotary joint that, in the rotary beam (RB) operating mode, is driven by a motor at 300 revolutions per minute (i.e. 5 Hz). Thus the radar beam is wobbled by a fraction (~0.2×) of the

50 Chapter 2: The Australian IMR Mini-Network beamwidth about the vertical axis. The transmitter has a peak power of 25 kW and transmits pulses of duration 0.05 µs at 2560 Hz, synchronised to the beam rotation, in the RB mode, or 2500 Hz in the stationary beam (SB) mode when the feed remains stationary.

The 8-channel gated peak-hold unit and one of the PC DAQ boards comprise the data-acquisition system for signal conditioning, sampling and digitising. The signals are captured simultaneously at 8 discrete altitude ranges in gates of 50 m depth and 150 m apart (starting at 200 m AGL) at the radar pulse repetition frequency (PRF), i.e. 2560 Hz in the RB operating mode and 2500 Hz in the SB operating mode. The IMR beam produces a horizontal resolution of a few to several tens of metres at the lower and higher boundaries of the detection range. This fine spatial resolution enables detection of individual nocturnal migrating insects, which have been estimated to typically have 30 to 150 m between them during a heavy migration (Riley & Reynolds 1986). The captured signals pass through low-pass anti-aliasing filters with a cut-off frequency of 128 Hz. The signals in each channel are then digitised at one-eighth of the primary sampling rate, i.e. 320 Hz and 312.5 Hz in the RB mode and the SB mode respectively, by the data-acquisition board, and stored in a multiplex format in the PC extended memory using the PC’s ISA (Industry Standard Architecture) buses. The sampling resolution is 12-bit, i.e. 2.44 mV per binary number in the voltage range of -5.0 – 5.0 V.

The commercially available AWS (Monitor Sensors Pty Ltd, Brendale, Qld, Australia, http://www.monitorsensors.com), is fitted with five digital sensors for measuring wind direction, wind speed, air temperature, humidity and air pressure. The sensors are connected to an 8-channel 32 KB data-logger (type GLX-32D), which is set to log sensed data at 6-min intervals. The whole system is powered by an external DC source supplied from the mains, with a rechargeable battery installed for backup during power failures. The data-logger is connected to the PC via an RS-232 serial communications port and can be reprogrammed remotely from the PC, which also downloads the logged data.

The IMR unit at Thargomindah is similar in configuration to that at Bourke with the difference that this unit has a Racal 65125 transceiver (Racal Decca Marine Ltd, Surrey, UK) which has a better noise figure and no swept gain, a 15-channel gated peak-hold signal conditioning and sampling unit with gate width of 25 m and an 2 The Mini-Network Structure and its Components 51 observing range of 175 – 1300 m AGL, and a 128 KB data-logger (GLX-128D) for the AWS (Figure 2.4).

Figure 2.4 The IMR Unit at Thargomindah Airport, Queensland This photo shows the air-conditioned cabin that hosts the IMR equipment and its control PC, the IMR antenna covered with a weatherproof radome, and the automatic weather station. The whole unit is secured within a fence. Photo courtesy of V A Drake, taken immediately after the IMR installation on 19 August 1999.

3 The Software Implementation

Automation has been a primary objective of the software design for the IMR units. To minimise the human interaction required to operate the radar, all equipment control and operation functions, data acquisition, data processing and archiving are implemented as autonomous procedures controlled through preset schedules and options; these all run without user interaction. Data processing, which includes the retrieval of target parameters from digitised radar signals and the statistical summary of observation results, is implemented entirely on the IMR control PCs. The primary analysis results are transferred to the base laboratory via the modem-telephone link. The digitised radar signals and the observation results are later archived to the base laboratory via the postal

52 Chapter 2: The Australian IMR Mini-Network service (Figure 2.5). Since system stability and reliability are essential for long term operations, all programs include routines to monitor hardware performance and to detect any system faults.

Lyhhmmss.ddd

yymmmdd.log yymmmddc.txt CD Recording Ryhhmmss.ddd yymmmdd.fit

Syhhmmss.ddd yymmmdd.wdt yymmmddc.gif

Processing 1 Processing 2 Wyhhmmss.ddd Postal Service DE.EXE FitDG.EXE Sampling DispWD.EXE FitHG.EXE FitPG.EXE VBIMRH.EXE Monitor.EXE

IMR, AWS IMR Control Computer Network Server Figure 2.5 Flow Chart of IMR Data Stream The solid lines indicate the process of signal digitisation, data analysis, and results transfer to the web server at the base laboratory. The program names and file types are shown on the chart. The dashed lines indicate the process of data archiving. yy, mmm, dd, hh, mm, ss and ddd in the file names represent, respectively, the year, month, day, hour, minute, second and Julian day when the files were created. c in the file names indicates the different contents of the result files: displacement directions, orientations, speeds, masses, wingbeat frequencies, etc. Figure adapted from Drake et al. (2002a).

The whole software system is based on the MS DOS (Version 6.20, Microsoft Corporation, Redmond, WA, USA, http://www.microsoft.com) platform, which is a matured, stable single-task operating system and does not require too many resources for installation and running. Its text-based user interface simplifies its use over a modem connection, while the single-task operating system makes control programming straightforward. The programming language is Borland C++ for DOS (Version 2.0, Borland International, Inc., Scotts Valley, CA, USA, http://www.borland.com), which combines the low-level access to system resources and hardware needed for equipment control with the high-level programming functions required for scientific calculations, graphics plotting and streamed file handling. Using this compiled language instead of an interpreted language results in fast execution speed because of the high efficiency of the 3 The Software Implementation 53 compiled code, and makes real-time operation practicable using relatively low-speed processing units. The stand-alone executable programs are invoked simply and complete their tasks without further interaction. The object-oriented modular programming language ensures portability to any future redeveloped system.

3.1 The IMR Operation

The program VBIMRH, which was developed by I. T. Harman, operates the IMR and digitises radar signals. It brings the radar into operation incrementally, with careful system testing and performance monitoring at each step to detect any hardware failures. Hardware outputs are fed back, mainly through the multi-function unit, for comparison with pre-tested ranges. Once initial procedures (including transmitter warm-up) are completed, VBIMRH runs the IMR on a periodic schedule in which observing and performance checking occur every hour. The schedule alternates operation between RB and SB modes. It shifts the radar gate ranges between sampling periods, so that the 8 discrete gates cover the whole 200 – 1400 m range of observation heights in a sequence of 3 continuous RB operating periods (Figure 2.6a). The SB mode operations are also made in sets of 3, with the beam rotated through 60° between them; this guarantees any collective orientation of a target population would fall within 30° of one of the three beam polarisations, so that at least some strong echoes will be obtained (Figure 2.6b).

500m E 400m N N N

300m C Z Z C Z 200m E C

100m 1 2 0m 3 E

123 (a) (b)

Figure 2.6 Schemes of Range Shifting (a) and Beam Positioning (b) Used by the Bourke IMR Samples in a set of 3 cover (a) the whole observing range of 200 – 1400 m AGL by shifting the 8 discrete 50-m-deep range gates (shown as shadings) in the RB mode, and (b) any possible collective orientation by rotating the off-set radar beam (centre at C) about the zenith (Z) and pointing the electric field (E, beam polarisation) to each of 3 directions 60° apart in the SB mode.

54 Chapter 2: The Australian IMR Mini-Network

Range shifting is not employed in the SB mode. The observations in each operation cycle are non-continuous as transmission is switched off briefly (for between 7 and ~100 s) during range shifts and mode changes. A rain detector and two temperature sensors are also controlled by this program. VBIMRH samples and digitises the radar signals, and writes them immediately to the PC’s hard disk.

VBIMRH writes two types of file: a log file Lyhhmmss.ddd in ASCII (American Standard Code for Information Interchange) format and binary-format files containing sampled data for the RB (Ryhhmmss.ddd) and the SB (Syhhmmss.ddd) observations. All files for a night are organised in one directory, the name of which is formed from the start date of the evening when the observations begin. The file naming system is described in Figure 2.5. The log file logs, for each observing cycle, the time, the tuning voltage for the tune point, the transceiver power, the eight peak-detector residuals (which indicate the system noise level), the equipment and outside temperatures, the motor current and the antenna alignment. It also logs, for each data sample, the motor speed (in RB mode), the radar PRF, the sampling frequency, the angular position of the parked antenna (in SB mode), the rain-detector reading, the eight gate ranges and widths, and the data-file name. It is updated at the end of each observation cycle to avoid data loss if the power fails or the system halts. The main electricity supply is switched off daily for a short period by a timer switch to reset the whole unit and clear any system halt that may have occurred.

The sampled signals are stored as 2-byte integers. Therefore, a sample file contains sequential records of 8 (Bourke IMR) or 15 (Thargomindah IMR) integers, each of which is a measurement of the signal power from one range gate, ordered from low to high altitudes. The records are continuous, without delimiters. Each file contains 49152 consecutive samples, and is therefore 786432 bytes (Bourke IMR, 8 channels) and 1474560 bytes (Thargomindah IMR, 15 channels) long and take about 154 s (RB mode) and 157 s (SB mode) to acquire. The schedule acquires 15 samples (6R, 3S, 6R; taking 39 min) each hour at Bourke, and 12 samples (6R, 3S, 3R; taking 31 min) each hour at Thargomindah. Operations are timed for 11 hr nightly, starting at sunset, and generate 124 MB (165 files) of data at Bourke and 186 MB (132 files) at Thargomindah each night. The fewer samples for each hour at Thargomindah are set for the sake of 650MB-CD data-archiving system. 3 The Software Implementation 55

3.2 The AWS Operation

The AWS as supplied provided weather observations only twice a day, at 9 am and 3 pm, and had to be reprogrammed for use with the IMRs. It comes with software which allows communication between the data-logger and the PC via an RS-232 connection, so that the logged data can be dumped to the PC and the data-logger reset. However, the supplied software requires user interaction, and cannot retrieve the logged data for a specific period. In addition, the software was not “Y2K-compatible” due to the usage of two-digit year representation. Therefore, a new program Monitor has been developed to operate the AWS from the PC and to obtain samples at 6-min intervals.

MONITOR bypasses the less-than-adequate PC BIOS (Basic Input/Output System) communications routines and installs a serial interrupt handler. Direct access to PC hardware allows fast transfer rates and eliminates the need to continuously poll the serial port for data. Data entering the serial port are stored in a circular buffer, thus eliminating data loss. MONITOR runs according to the settings in a configuration file, which specifies the transfer rate (which should be the same as the hardware setting on the data-logger), the serial port number, the logger channel types and the types of sensors connected, data calibration formulas for the conversion of the logged raw data into physical quantities, the LCD display formats, the starting time, the logging time interval, the end time, the data directory name, and the station elevation (which is needed for the calibration of air pressure). MONITOR runs with a command line parameter that specifies its operation: initialising the data-logger, uploading logged data to the PC (followed by clearing of the logger’s memory), acting as a terminal for diagnosis of any data-logger faults, or displaying and modifying the configuration file with prompts. The configuration file can also be edited with a text editor.

MONITOR writes two data files into the PC hard disk when uploading logged data. It waits until the end of scheduled data-logging time and initialises uploading of the logged raw data (at 9600 baud) to the PC, where the data are temporarily stored in an ASCII-format file (about 20 KB). It then cleans up the temporary file by removing extra CR characters (Carriage Return, ASCII 0DH) sent occasionally by the data-logger, and moves it to the IMR data directory for the previous night’s observations, with name Kyhhmmss.ddd. A binary file Wyhhmmss.ddd (~5 KB) is created from this raw data according to the conversion formulas. MONITOR then clears the data-logger memory

56 Chapter 2: The Australian IMR Mini-Network and waits for the next logging time to re-initialise the data-logger, as the data-logger must be reset at a time when its base period will divide exactly into 24 hr, otherwise it will not log any observations. The AWS has been set to log wind direction, wind speed, temperature, humidity and air pressure at 6-min intervals. Thus weather observations in one base period (6 min) are lost each time (usually once per day) that data are uploaded from the data-logger. In case MONITOR is not launched, e.g. if the PC halts from a system fault or there is a power failure, the logged data can be kept for up to 4.7 days at Bourke (GLX-32D logger) and 23.7 days at Thargomindah (GLX-128D). If overflow occurs, the earliest stored data are lost as the logger uses a circular memory.

A program DispWD reads the binary-format file Wyhhmmss.ddd and converts it to an ASCII-format file yymmmdd.WDT, which is used to generate the summary graphics of daily observations.

3.3 Extraction of Target Parameters

The program DE has been developed to automatically analyse the digitised radar signals and extract the flight and identity parameters of the detected targets. DE is based on the prototype IMR signal-analysis program FIT developed by I. T. Harman. DE uses the IMR log file Lyhhmmss.ddd as an index to Ryhhmmss.ddd and Syhhmmss.ddd data files. The program first extracts the file names for each sample from the log file, and also reads the IMR operation parameters (the sample type and the sampling altitudes). It then analyses each data file, or skips it if the rain detector indicates that it would be contaminated by rain. DE delimits and analyses each echo, estimating trajectory direction, body alignment, and radar cross-section (RCS) size and shape parameters (for RB-mode data), lower limit of the RCS (for SB-mode data), target speed, wingbeat frequency and intensity, and wingbeat harmonic intensities. Details of the parameter- extraction algorithms coded into DE are given in Chapter 3.

DE writes two ASCII-format files to the hard disk: a log file yymmmdd.LOG and a data file yymmmdd.FIT (Figure 2.5). The LOG file records the observation type (R – RB mode, S – SB mode) and rain status (r – raining, c – clear) of every sample in each observation cycle, the cycles being separated with CR/LF characters (Carriage 3 The Software Implementation 57

Return/Line Feed, ASCII 0DH and 0AH), and the total count of targets for the whole night. It is at most 406 bytes, for 15 samples each hour over 11 hours.

The FIT file records extracted target parameters from every delimited echo within each sample, results for the whole night being recorded sequentially (Table 2-2). This file has variable-length records ending with CR/LF characters. Each observation cycle is started with a record containing a single hash mark (ASCII 23H). Each sample is marked with its file name as the first record, followed by records of delimited echoes in sequence from sample beginning to end and from low to high gate range (i.e. altitude). The minimum length of echoes that are analysed is 0.8 s (256 sample points in 4 beam-rotation cycles) in the RB mode and 0.4096 s (128 sampling points) in the SB mode, while echoes of 0.6 s in the RB mode and 0.3072 s in the SB mode are counted but not analysed. The record for a short echo is given a code of -2. The record for an analysed echo has the estimated parameters in a fixed sequence delimited with spaces (ASCII 20H). If a quantity cannot be estimated, a code of -1 is substituted. The total count of delimited echoes is also recorded at the end of the file. Both the LOG file and the FIT file are placed in the data directory for the night, along with the original raw data. The biggest FIT files are about 8 MB (more than 54,000 targets) from the IMR at Bourke and about 13 MB (more than 95,000 targets) from the IMR at Thargomindah.

Table 2-2 Contents of FIT file Record type Record contents Units Observation cycle Hash mark (#) Sample File name [indicating date, time, and type (R/S)] Transect parameters (p, τ, φ) BW, s, ° R echo Displacement direction ° Body alignment (orientation) ° 2 RCS parameters ( a0 , a2 , a4 ) cm S echo Lower limit of RCS cm2 Altitude (centre of range gate) m Speed m/s Uncertainty of speed m/s Echo length s R & S echoes Spectral relative energy (entire) - Wingbeat frequency (fundamental) Hz Wingbeat relative energy (fundamental) -

Number of harmonics ( Nh ) -

[Harmonics (order, frequency, energy)] × Nh -, Hz, - Goodness of fit parameters ( r 2 , δ, etc.) various End of file Total count of targets -

58 Chapter 2: The Australian IMR Mini-Network

3.4 Statistics of Observation Results

The statistical and graphical summaries of nightly observations are generated by the programs FitDG, FitHG and FitPG, which produce time-series profiles and histograms of target numbers, flight parameters and identity characters. Relative target numbers have been preferred to the absolute aerial density of targets, which requires a relatively complicated calculation to estimate (Riley 1978; Drake 1981a; Chapman et al. 2002a). The results are produced in three forms: as tabular ASCII-format files, as graphics files in GIF (Graphics Interchange Format, CompuServe Interactive Services, Inc., Jacksonville, FL, USA, http://www.compuserve.com) format, and as graphics files in PS (PostScript, Adobe Systems Inc., San Jose, CA, USA, http://www.adobe.com) format. The GIF files are for screen viewing, and the PS files for a printed record. The weather observations from the AWS are also summarised.

All GIF graphs can be generated in scalable sizes with command-line options. The total graphic size of one night’s observations, with the options set to fit on a web page designed for 800 × 600-pixel screen resolution, is about 223 KB. Direct access to the video memory on a VESA (Video Electronics Standards Association, http://www.vesa.org) standard graphics card, or to the extended memory, enables fast generation of the graphics. All PS graphs are designed for printing on A4 paper, but are not generated in daily operations.

The time-series profiles of target numbers, flight speeds, displacement directions, orientation directions, masses and wingbeat frequencies are generated by FitDG (Table 2-3). This program reads the FIT file (using the LOG file as an index), sums target numbers, and averages each of the parameters. Statistics are calculated for eight 150-m-deep altitude ranges for each series of 3 continuous samples, and stored in an ASCII-format file. The total counts of all detected targets and counts of well estimated targets (good quality echoes, see Section 3.1 of Chapter 3) are plotted as bar charts, while the averages of various parameters are plotted as dashed lines. Periods of rain recorded by the rain detector are shown as shading in the background of every graph. FitDG also plots the surface weather – wind direction and speed, temperature and humidity – for the same period as the IMR observations, using the WDT file from the AWS observations as input. Figure 2.7 and Figure 2.8 are example outputs, from an 3 The Software Implementation 59 occasion when a massive immigration of Australian plague locust was detected after a wind change. Most of the graphs are less than 30 KB in size.

Table 2-3 Input and Output Files of FitDG File Contents Size (KB) Input Files (all ASCII format) yymmmdd.LOG IMR observation log <0.4 yymmmdd.FIT IMR observation results variable yymmmdd.WDT AWS surface weather observations 9.7 Output Files (ASCII and binary formats) yymmmdd.TXT Time-series profiles of parameter averages variable yymmmddN.GIF|PS Time-series profiles of target numbers 39, 8 yymmmddD.GIF|PS Time-series profiles of displacements 27, 8 yymmmddO.GIF|PS Time-series profiles of orientations 27, 8 yymmmddS.GIF|PS Time-series profiles of speeds 27, 8 yymmmddM.GIF|PS Time-series profiles of masses 27, 8 yymmmddF.GIF|PS Time-series profiles of wingbeat frequencies 26, 6 yymmmddW.GIF|PS Surface time series of wind, temperature, humidity 25, 7

Figure 2.7 Time-Series Profiles of Target Numbers (a) and Wingbeat Frequencies (b) at Bourke during the Night 03-04 Dec 1999 A massive immigration of Australian plague locusts was detected after a wind change occurred at 21:40 h Australian East Standard Time (see Figure 2.8b). Counts from RB and SB modes are shown in different colours, as are the total counts and the counts of good-quality targets (see Section 3.1 of Chapter 3 for criteria). Other estimated parameters are calculated as the averages over a period of sampling time for a certain altitude range and plotted as dash lines according to the scales of time and altitude. Graphics generated by FitDG in GIF format.

60 Chapter 2: The Australian IMR Mini-Network

Figure 2.8 Time-Series Profiles of Displacement Directions (a) and Surface Weather Conditions (b) at Bourke during the Night of 03-04 Dec 1999 The averages of displacement directions are calculated in degrees by vectorial addition over a period of sampling time for a range of altitude and plotted as dash lines in the vertical scale of degree. The wind directions (deg), wind speeds (m/s), temperatures (ºC) and relative humidity (%) were observed at 6-min intervals by the automatic weather station installed with the IMR. A wind change occurred at about 21:40 h AEST. Graphics generated by FitDG in GIF format.

The histograms of flight parameters and identity characters are generated by program FitHG, again using the LOG and FIT files as input. The calculation results are stored in the ASCII-format files yymmmddc.TXT (c: D – Displacement direction, O – Orientation direction, S – Speed, M – Mass, F – wingbeat Frequency) which contain quantities from well estimated echoes, and the graphics file yymmmddH.GIF (~6 KB) or yymmmddH.PS (~7 KB) which presents full-night histograms of the principal parameters in a single graphic. Equal-area rose diagrams (Batschelet 1981; Baas 2000) are used for target displacement and orientation directions using a 10° bin sector width and bar histograms for target speeds, wingbeat frequencies and masses (plotted on a logarithmic scale). Some count totals and observation summaries are also printed on the graphic. Figure 2.9 shows an example of this histogram graphic, for another night on which Australian plague locusts were migrating in large numbers. 3 The Software Implementation 61

Figure 2.9 Histograms of Target Parameters at Bourke for the Night of 26-27 Feb 2000 The target parameters of displacement direction (a), orientation direction (b), speed (c), wingbeat frequency (d) and mass (e) are shown as equal-area rose diagrams and bar charts. The histograms of wingbeat frequency and mass are plotted in one graph with overlays in a different colour. The scales for wingbeat frequency are on top and right axes and for mass on the bottom of logarithmic scale and left axis. The total count of 13174 targets includes short and unanalysed echoes. The well estimated targets comprise 32% and 16% of successfully analysed echoes for the RB and SB modes respectively. Graphics generated by FitHG in GIF format.

The program FitPG summarises the observations for the whole night from the FIT, LOG and WDT files, generating graphics representing the size of the airborne population, accumulated hourly target profiles and time-series of surface weather conditions. An index graphic of insect intensity, yymmmddI.GIF (< 300 Bytes), is generated as a vertical bar showing the average number of targets in 30 min of observation, normalised to a maximum of 4000. The graphics yymmmdd1.GIF and yymmmdd2.GIF (< 400 Bytes) are larger, horizontal versions of the index bar, and show the starting date of the observations and the size of the airborne population in colour. The time series of hourly profiles is an animated graphic in a GIF file (yymmmddP.GIF,

62 Chapter 2: The Australian IMR Mini-Network

~20 KB) or a stack-bar plot in PostScript format (yymmmddP.PS, ~4 KB). The animated GIF file is also printable as its first frame shows the accumulation of the whole night’s observations. Due to the different echo selection criteria adopted for analysis of RB- mode and SB-mode signals, only targets detected in the RB mode are plotted on this graphic. The weather time series appear on a single graphic yymmmddC.GIF (~4 KB), which is the condensed version of yymmmddW.GIF. Figure 2.10 shows the graphics produced by FitPG for the night from which Figure 2.9 was produced.

Figure 2.10 Indices (a, c) and Profiles (b) of Target Numbers and Condensed Summary of Surface Weather (d) at Bourke for the Night of 26-27 Feb 2000 The profile of target numbers is animated to show how the profile accumulates at each observation cycle (i.e. 1-h intervals). Graphics generated by FitPG.

4 System Operation and Network Communications

The daily radar operation, on-line publishing of observation results, and archiving of the raw data and the processed results are now described. This section also gives an account of how the system is maintained.

4.1 Scheduling of Tasks

All IMR and AWS observation tasks, plus data processing and archiving, are conducted by individual stand-alone programs that are launched by a schedule program, CronJr (Version 2.55e, Software Shorts, California, USA). This method was selected following consideration of a number of alternatives, none of which would have been able to provide all the required functions. These methods were:- 4 System Operation and Network Communications 63

a) A big batch file. This is the simplest choice, with tasks being carried out in a sequence. However, it cannot meet the requirement that programs run at specific times, and it cannot skip a task that has become overdue. b) A foreground program to load each task as a child process. This would occupy too much memory, and memory is critical on the MS-DOS platform. In addition, if a task is conducted by several programs in a batch file, another copy of COMMAND.COM needs to be loaded on top of the foreground program, which further reduces the available memory. c) A small TSR (Terminate and Stay Resident) program. This could schedule tasks, but it takes clock cycles and substitutes interrupt vectors and this may cause conflicts with other programs. It will also occupy some memory. In the method chosen, two levels of operating system batch file are used, one to update the schedule and the second to conduct tasks as they fall due. The first-level batch file calls a foreground program, which checks the schedule against the system clock and composes a second-level batch file to execute the due task. The foreground program terminates itself when a task becomes due, and the second-level batch file is launched by the first-level batch file. Upon completion of each task, the second-level batch file unloads itself from memory and the first-level batch file reloads itself with the updated schedule to replace the copy in memory, and the cycle restarts.

The IMR is currently scheduled to observe for 11 h every night, starting at sunset. As soon as the IMR observations terminate at dawn, the surface weather data monitored by the AWS over the previous 24 hr is downloaded from the data-logger to the PC’s hard disk. The primary data analysis is then launched, and when this is completed the statistical summaries and graphics are generated. The results are normally available by late morning. The GIF graphics files are downloaded routinely to the network server, to allow immediate dissemination via the WWW. Downloading is currently done manually; automatically uploading from the IMR control PC to the web server is feasible, but has not been adopted. Data archiving is scheduled for the early afternoon. The raw data and processed results are burnt to CDs and the occupied disk space is freed only after the data archiving is reported successful by the CD burner. The tasks of changing CDs and mailing them to the base laboratory are performed by a local agent, approximately once (at Bourke, where 5 nights of data fit onto a CD) or twice (at Thargomindah, 3 nights per CD) a week.

64 Chapter 2: The Australian IMR Mini-Network

4.2 Integration and Dissemination of the Observations

On-line publication of the IMR observations occurs via a web site. The design of the web site is based on considerations of easy maintenance, economics, portability and security, and ease of browsing with minimal user interactions. Four possible implementation plans were investigated:-

a) Generating a web page manually every day, probably by modifying a template.

b) Generating a web page automatically, by a program running on the server, according to a schedule and contingent upon data availability.

c) Generating a web page instantly, by a CGI (Common Gateway Interface) script or program, when required by a user.

d) Generating a web page, depending on the user’s browser, instantly on-the-fly (i.e. without it being saved on the server) by a client-side JavaScript that interacts with a Java applet.

The last seemed the most satisfactory approach, as it does not require further action once the observation results are stored on the web server. It also saves web space since the only files stored permanently are the original GIF graphs, and it does not require additional server resources, such as database or CGI program support. In addition, it avoids the security risk of allowing file-writing on the server, as JavaScript is a client- side language and the Java applet is also a client-side program which can only access the web server on which it is hosted. Furthermore, this approach makes the web site transportable as both JavaScript and Java are platform-independent. However, a number of programming challenges had to be overcome in the implementation of this Java- JavaScript solution. These are described in Appendix A.

The web site has two levels of pages. The front page (Figure 2.11) summarises one night’s observations. It includes the histograms of displacement direction, orientation, speed, mass and wingbeat frequency, the animated profile of hourly target numbers, and the condensed summary of surface weather conditions. Each of these can be clicked to load one of the second-level pages, which present more detailed summaries for each parameter (e.g. Figure 2.7 and Figure 2.8). The front page also indicates the day-to-day changes in the level of insect activity, by means of bar charts at left and at the bottom of the page, for the past two weeks and four months (“season’s 4 System Operation and Network Communications 65 dynamics”) respectively. The index bars at left can be clicked to load a similar page for the indicated date, and the bars at the bottom can be used to identify the dates by holding the mouse point over them. A help icon (question mark) will display on the screen at top-right for 10 s each time the mouse is moved. If clicked, it toggles the appearance of an explanation of each summary graph that pops up when the mouse is held over the graph (Figure 2.12). These explanation panels may be helpful to new users but will be annoying to regular visitors, and have therefore been made optional. Two URL links are also given, for connections to The Radar Entomology Web Site (http://www.pems.adfa.edu.au/~adrake/trews/) at the bottom left, and a set of pages with background information on Insect Monitoring Radars (http://www.pems.adfa.edu.au/ ~adrake/imr/ww_imr_hp.htm) at the bottom right, both are maintained by V. A. Drake.

Figure 2.11 Front Page of the IMR Web Site of Bourke This is the summary page of IMR observations at Bourke for the night of 25-26 September 1999, when two layer concentrations were present. Periods of heavy migration during the previous 4 months can be identified from the interactive bar charts at left and at the bottom of the page.

66 Chapter 2: The Australian IMR Mini-Network

Figure 2.12 Front Page of the IMR Web Site for Bourke, Showing a Pop-up Explanation An explanation note appears when the help icon (question mark at top-right) is switched on. Every graph on the front page has a short explanation, intended primarily for new users.

The major delay on visiting this web site is waiting for the client-side Java applet to check the data availability with the web server. It takes about 2.5 min for the front page (~ 170 small or tiny GIF files yymmmddI.GIF, ~ 102 KB in total) to be loaded over a modem connection at 33.6K, or 20 s on the campus LAN. The secondary pages are individual GIF files (<30 KB), and each takes less than 10 s to load over a modem connection.

4.3 Data Archiving

The raw data and processed results are fully archived on CDs. Commercially available software package DAO (Version 3.5 for 16-bit DOS, Goldenhawk Corporation, Milford, NH, USA, http://www.goldenhawk.com) is used to create CD images (ISO 9660 format) and to burn them on to CDs. DAO is a set of stand-alone programs to handle different tasks related to CD copying and writing. These programs can be run 4 System Operation and Network Communications 67 interactively with user confirmation on every step, or silently with command line parameters. They return standard error codes when they terminate, and therefore can be called from a batch file or from another program that incorporates error-checking routines.

Data archiving has been implemented as a fully automatic procedure. Two programs have been written in Borland C++, DataTrC and DataTrB, which call DAO programs for, respectively, creating and burning CD images. DataTrC converts the data to be archived into a CD image. It ensures that the available data are sufficient to nearly fill a CD, and that a partition with enough free space for saving the CD image is available. It then moves an appropriate amount of data to a temporary directory and calls MakeISO from the DAO package to convert the whole directory recursively into a CD image. The temporary directory is deleted once the CD image has been created successfully. For the convenience of future analysis, data directories, each of which contains one night’s observations, are not split. DataTrB conducts the task of burning a CD image to a blank CD. It searches for a CD image and checks if a blank CD is available in the CD burner, then calls DAO from the DAO package to burn the image. Upon successful CD writing, the CD image is deleted to free the occupied space. All activities are logged for inspection.

4.4 Routine Inspection and Maintenance Servicing

The IMR network has been developed to maximise automation, to incorporate a self- diagnosis capability, to minimise operator workload, and to minimise the need for maintenance visits to the field sites. However, routine inspection of system logs can detect hardware faults, and some fault diagnosis can be done remotely with or without help from the local agents. On-site servicing, nonetheless, is required when major faults develop or the hardware is to be upgraded.

A background communications package, TeleReplica (Version 4.37, D. Thomson, Monash University, Melbourne, Vic, Australia, http://www.gscit.monash. edu.au/~dougt/telereplica.html), has been used for remote access to the IMR-control PCs. The software can run in an MS-DOS prompt under MS Windows, or a DOS emulation (DOSEMU, http://www.dosemu.org) if the server is running Linux. Server

68 Chapter 2: The Australian IMR Mini-Network and client applications have been installed on both the IMR-control PCs and the network server. A 10 KB TSR program TRHost acts as the host and monitors a specified serial port for incoming calls. It can be set running at both ends so that communication can be initialised in either direction. The program TRDial (or TR if TRHost is not in memory) on a remote PC can then dial into the host; the remote PC then takes the host computer over, acting as its text terminal. File transfer can be conducted in either direction once a connection is made. The host resident program is loaded into memory by the DOS automatic batch file Autoexec.BAT on the IMR- control PCs to ensure they are always accessible. TeleReplica runs according to an identical configuration file at both the host and the remote ends for automatic dial-up, synchronisation of baud rate and password check for authorisation. Therefore, dial-up and login is an automated and secure procedure. The automatically transmitted password can be set as long as 39 characters for high security. TRHost can be removed when the IMR operation starts, but as it has been confirmed not to conflict with VBIMRH or any other software, it has been kept running in background so that the IMR operation can be monitored or controlled in real-time if desired.

Regular servicing is done remotely from the base laboratory via the modem and telephone connection. Inspection of the IMR’s performance and downloading of the observation results are conducted daily during the locust season, and at least two or three times a week at other times. Checking daily operation logs, weather observations, and the data-archiving log enables a quick determination of whether the hardware and software have worked properly. System settings and operation procedures can be modified remotely, by changing either program configurations or launching schedules. Software upgrades and bug fixing are also done remotely, either modifying and recompiling source codes on the IMR control PC or uploading newly compiled versions. System diagnosis to identify hardware faults, and implementation of fault work-arounds, is mostly undertaken remotely, e.g. by running the IMR manually to check the hardware’s responses at each step and function, or running Monitor in terminal mode to check the function of each AWS sensor.

On-site servicing is undertaken by a local agent and during visits by the base staff. The local agent makes regular inspections of the IMR site for any obvious damage or threat, changes CDs of archived data and posts them to the base laboratory, and 4 System Operation and Network Communications 69 liaises with local authorities, contractors and public. The local agent can also help by visually identifying obvious faults and solving minor problems, such as physical damage and cuts in the power supply or telephone line, and may be able to dismount some hardware items and dispatch them to the base laboratory for repair. However, a servicing visit is unavoidable when major hardware maintenance is required. A servicing visit to both IMRs takes 5–6 days for the staff from the base laboratory, of which 3 days are spent travelling to and from the sites by road. Nevertheless, the IMRs have proven very reliable during long-term operations, and 2–3 servicing trips per year have sufficed.

5 System Performance

The IMR observation results have indicated that both the system hardware and the software function well. Some observation results are presented here to demonstrate the data quality.

5.1 System Reliability

The IMR at Bourke had achieved successful observations on more than 85% of its scheduled observing periods between the commencement of its operation on 28 April 1998 and its first serious failure on 04 November 2001 (1116 out of 1286 nights). The main causes of data loss are summarised in Table 2-4. System faults caused the most losses. However, most of those occurred in the early stage of operation when the radar software was still undergoing some development and only a small-capacity hard disk was installed. CD archiving failure and losses in or after delivery caused a quarter of all data loss. On-site servicing, on the other hand, had only a minor impact on IMR operation.

The IMR at Thargomindah operated successfully on more than 88% of nights between 01 Sep 1999 and its first major failure on 16 Jan 2001 (447 out of 504 nights). Power failures, which often occur during thunderstorms at this site, were the major cause of data loss.

70 Chapter 2: The Australian IMR Mini-Network

Table 2-4 Causes of Data Loss of the IMR at Bourke Data loss Percentage of Percentage of Cause (nights) total loss (%) total nights (%)

System faults (disk full, hardware and 44 25.9 3.4 software faults) CD archiving and posted losses 41 24.1 3.2 Power failures 33 19.4 2.6 Suspension during servicing 21 12.4 1.6 Other or unidentified reasons 31 18.2 2.4

Total 170 100% 13.2%

5.2 System Utilisation

The IMR network has shown its potential for the study of insect migration and airborne fauna. Insect target identification, flight characteristics, and migration behaviour can be inferred or analysed from the IMR observations (see Chapters 4, 5 and 6). Some data obtained from the IMR observations are shown below.

8000

6000

4000

2000

0 Target Number of 20-21h Observations of 20-21h Number Target 9 9 0 1 -98 9 00 -0 -00 v b- y v ug- a -No -May-99 -A -Fe M -No -May-01 9-Aug-98 7 6 0 6 5 2 2 25-Feb-99 2 24 22-Nov-9 2 20- 18-Aug-00 1 14-Feb-0 1 Date (1-Jun-98 ~ 31-May-01, 933 nights, Bourke) Figure 2.13 Time Series of Target Numbers at Bourke for the Period June 1998 to May 2001 Counts are counted echoes observed, at all heights, between 20:00 and ~21:40 h AEST each evening, and that were analysed successfully.

The seasonal dynamics of insect abundance, which can be inferred from the IMR daily observations, provide an example of the IMR’s usefulness. Figure 2.13 5 System Performance 71 shows the numbers of insects detected in two hours of observations each evening at Bourke during the three years from June 1998 to May 2001. It is obvious that insect abundance has an annual cycle, in which numbers increase in spring to a summer maximum and decrease in autumn to a winter minimum. This reflects the seasonal changes in temperature and is consistent with general entomological knowledge of insect seasonality (Wolda 1988). In addition, there are differences between years which presumably relate to changes in the general suitability of conditions for insect development (which in this region probably mainly reflects variations in rainfall), with the year 1999–2000 having fewer insects than the other two years.

The equivalent time series from Thargomindah also shows a seasonal pattern (Figure 2.14), though it appears somewhat less extreme here than at Bourke. This is probably due to the climate differences between these two locations, with Thargomindah more clearly in the subtropical zone than Bourke (Tapper & Hurry 1993). Temperature in Thargomidah is generally higher than in Bourke and therefore many migratory insects can possibly get airborne most time when the local temperature is higher than their temperature thresholds of take-off.

12000

10000

8000

6000

4000

2000 Target Number of 20-21h Observations Numberof 20-21h Target 0 9 0 0 0 0 0 0 99 9 99 0 0 00 0 0 0 00 t- t- - - -00 t- c v-99 c n-00 pr- n- p Oc o a A Ju ug Oc - -N De -Feb-0 - -A -Se Dec- 0- 8- 8-May-00 27-Jul- 6 5- 4- 01 31-O 30 3 29-J 28 29-Mar- 2 2 27 2 25 2 24-Nov-00 2 Date (2-Sep-99 ~ 16-Jan-01, 424 nights, Thargomindah) Figure 2.14 Time Series of Target Number at Thargomindah for the Period September 1999 to January 2001 Target counts are in the same way as in Figure 2.13.

The spatial pattern of insect movements can also be monitored by the IMRs. As an example, Figure 2.15 shows histograms of the nightly averages of insect displacement directions at intermediate heights at the two sites over the 16-month

72 Chapter 2: The Australian IMR Mini-Network period when both IMRs were in operation. These start to provide an indication of the general trends of population movements at these sites.

(a) Bourke 0 (b) Thargomindah 0 Total: 310 Total: 343 (500-650m) (475-625m)

270 90 270 90

180 180 Figure 2.15 Polar Histograms of Insect Displacement at Bourke (a) and Thargomindah (b) The equal-area rose diagrams for the mean direction of insect displacements, when they were concentrated significantly (P < 0.05 in Rayleigh test), for the observations between 20:00 and ~21:40 h AEST at 500-650 m above ground level at Bourke (a), and between 20:00 and ~21:30 h AEST at 475-625 m above ground level at Thargomindah (b), over the same period 2 September 1999 to 16 January 2001.

6 Discussion

Although the Australian IMR network is of minimum size, it has demonstrated its potential for long-term monitoring of airborne insect fauna and key migratory insect pests. Based on the knowledge gained from past research, carefully selected IMR sites on the main migration routes of economically important species will be able to play an important role in monitoring and forecasting (Drake et al. 2001). The current two IMRs at Bourke and Thargomindah have shown that major movements of the Australian plague locust can be detected (see Chapter 5). The IMR observations also help to establish the migration pathways of key migrants and have already provided stimulus for future research.

The simplicity and reliability of the IMRs have greatly aided their application to practical pest monitoring. The IMR network is an almost fully automatic system, from radar operation through to presentation of the observation summaries. The IMR hardware is much simpler than that of traditional scanning entomological radars, and the 5 System Performance 73 operating procedures much less dependent on the availability of skilled personnel (Riley et al. 1992a; Drake 1993). A low-end personal computer is sufficient for all required operation controls, data acquisition and analysis. In addition, the telephone link is economical and easy to set up, but meets the communications requirements for software maintenance, system monitoring and results transmission. Hence the construction and running costs are greatly reduced in comparison with manually operated scanning radars.

The software implementation has also achieved high efficiency and reliability. The DOS platform reduces the software complexity and the requirement for computer resources. All software for system control, data acquisition and processing is coded with C/C++, which is a powerful high-level language that is easy to maintain and transfer to alternative platforms. Online dissemination of the observation results has similarly been implemented in a platform-independent way, using Java and JavaScript languages on a standard HTTP web server. Self-development of software has reduced the system cost and allowed continuous improvement during the period of operation. Automation allows the system to run routinely without the need for staff with a high level of technical expertise. The nearly real-time delivery of observation results on the Internet enables users to promptly access them in a readily interpretable form, and to use them to assist decision-making in the practice of preventive control (Deveson & Hunter 2002; Hunter & Deveson 2002).

As will be discussed in later chapters, this IMR mini-network can be used to identify and monitor migrations of key migratory insect pests, such as the Australian plague locust, between the arid inland and the agricultural and horticultural regions of eastern Australia, and to explore their migration pathways and behaviours.

3 Implementation of IMR Signal Processing

1 ALGORITHM OF ECHO DELIMITATION...... 76 2 PARAMETER EXTRACTION...... 79 2.1 PARAMETER EXTRACTION FROM A STATIONARY-BEAM ECHO...... 81 2.1.1 Signal Composition of Stationary-Beam Echo...... 81 2.1.2 Parameterisation of Stationary-Beam Echo...... 82 2.1.3 Implementation Optimisation of Spectral Analysis ...... 88 2.2 PARAMETER EXTRACTION FROM A ROTARY-BEAM ECHO ...... 92 2.2.1 Signal Composition of Rotary-Beam Echo...... 93 2.2.2 Parameterisation of Rotary-Beam Echo ...... 95 2.2.3 Improvements of Computation ...... 104 3 SOME MEASURES OF IMR PERFORMANCE ...... 105 3.1 PERFORMANCE OF ECHO-DELIMITATION ALGORITHM ...... 105 3.1.1 Algorithm sensitivity ...... 107 3.1.2 Algorithm reliability...... 109 3.2 MEASUREMENT ACCURACY ...... 111 3.2.1 Angular range of detection...... 111 3.2.2 Body alignment of target...... 111 3.2.3 Target speed...... 112 3.2.4 Wingbeat frequency...... 114 4 CONCLUSIONS...... 115

Analysis of entomological radar data has always been time-consuming, and this has restricted its application in the practice of pest management (see Chapter 1). To speed up the data processing, an automated procedure had to be implemented, for selecting target echoes from the radar signal and retrieving target information from them. The program DE, which has been introduced in Chapter 2, has achieved this objective and is described in detail with examples here. In comparison with its prototype program FIT, DE has been improved with a new algorithm to enhance accuracy and efficiency in echo delimitation, and rewritten codes of least-squares fit and fast Fourier transform (FFT) incorporating weighting functions to improve the quality of parameter estimation and the stability and speed of processing. Part of this work, the implementation of new FFT spectral analysis and weighted least-squares fit and methodology for retrieving wingbeat

75 76 Chapter 3: Implementation of IMR Signal Processing frequency and its modulation parameters from rotary-beam signals, has been published in Drake et al. (2002b) and Wang & Drake (2004).

1 Algorithm of Echo Delimitation

Identifying and separating target echoes from background noise and adjoining echoes is essential for the processing of digitised radar signals. Only an automated procedure will be able to cope with the large amount of data accumulated from long-term IMR observations. The echo-delimitation algorithm in the prototype program FIT is based on threshold tests of signal intensity and its variance level. This appears to be effective when aerial densities are low, but fails on nights of high migration intensity as the signal strength then often remains continuously above the threshold while echo boundaries are not clear due to interference from nearby targets. Hence, the selected signals have often reached the maximum acceptable length, spreading over a number of targets which cannot be identified and analysed. Although the FIT algorithm incorporates a secondary delimitation process that does a simple sign test of intensity variation on the selected signal, inspections of the data and the algorithm outputs showed that at high densities the echoes were hardly ever selected properly. Thus, target densities were often estimated unreasonably low during heavy migration nights, especially at short ranges. In addition, assuming the same background noise level for all peak-detector gates produces many “bottoming out” echoes, i.e. the background noise is selected as

pMark echo1 echo2 echo3 echo4 echo5 (a) pVal (b) rMark 6dB cM lMark rVal Vt Ph lVal Th lPospPos rPos Figure 3.1 Criteria of echo selection (a) and possible echoes selected (b) Diagram (a) shows how a strong echo is delimited from adjoining echoes and (b) shows how five adjoining echoes might be delimited. Periodic averages of the signal strength cM are plotted on this graph. The selected echo is shown with three boundary marks, which indicate their signal positions and strengths. Key: l – left, p – peak, r – right, Th – threshold height (2 dB above the minimum value of the periodic averages), Ph – minimum analysable strength (6 dB), Vt – variation tolerance (1.5 dB), Mark – position label, Pos – position, Val – signal intensity. 1 Algorithm of Echo Delimitation 77 part of the echo. Further more, the two steps of echo selection consumed a lot of computation time. Therefore, an alternative algorithm that delimits echoes both precisely and efficiently had to be developed.

The new echo-delimitation algorithm employs a one-way peak-finding and shape-checking scheme. As the received signal is expected to exhibit a parabolic variation of power intensity when a target passes through the radar beam, because of the Gaussian beam shape and the logarithmic receiver characteristic (Drake et al. 2002b), an increase of signal strength indicates the possible start of an echo. The first calculation step is to average the time series of digitised radar signal from each range gate over each scan cycle (64 samples, 0.2 s) for the rotary-beam (RB) signals, or over 32 consecutive samples (~0.1 s) for data from the stationary-beam (SB) observations. This eliminates the modulations from the scan and smooths any wingbeats. A shorter averaging period, e.g. the half scan cycle, which would provide improved echo separation especially on samples from nights of heavy migration, is also possible for the RB signals. However, this will make the echo process much complicated. Therefore, the short averaging period is only adopted for SB data. Next, the averaged signal strengths, cM, are examined for this probable parabolic form. If the cM value is stronger than the threshold Th, which is 2 dB above the higher value of the expected background noise and the minimum value of the averaged signals (Figure 3.1a), the signal intensity will be stored in lVal, its position in lPos, and the left boundary lMark of the echo will be marked. These variables will be continuously updated if the signal goes weaker, or discarded if the signal goes below Th. When the signal is getting stronger than the analysable strength, i.e. 6 dB (Ph) above Th, its strength is saved into pVal, its position in pPos, and the echo peak pMark labelled. The peak position and its strength will be replaced by the new values if the signal goes even stronger. If the signal strength goes down more than the required strength difference of Ph from its peak position, the echo right boundary rMark will be labelled, and its position and value are recorded in rPos and rVal, indicating an echo selection. Again, the right boundary will be renewed if the signal amplitude continues going down. However, if the signal strength passes the level of Th, or increases again, e.g. due to a second target, by more than a tolerance Vt of 1.5 dB, delimiting of this echo is terminated. The left boundary lPos and right boundary rPos of this identified echo are returned as indices to their corresponding piece of raw signal for parameter retrieval. The algorithm flowchart is shown in Figure 3.2.

78 Chapter 3: Implementation of IMR Signal Processing

Begin

N End Read in Y More N Echo Got ech o? next cM signal? Y processing

cM-preV? N >Vt

Y Y Y

Y N N N N rMark ? Echo>xL? cM>preV? cM

N Y

pVal-cM Y >Ph ? √rMark √rPos, rVal

N Y N Y N YY√pPos, pMark ? Echo>xL? cM>preV? cM>pVal? pVal

N N

Y N N N lMark ? Echo>xL? cMpreV? √lPos, lVal

N YYY

N N cM>Th ? cM>Ph?

Y Y

√lMark √pMark ×lMark √lPos, lVal √pPos, pVal

Y pMark cM>Ph ? √ √pPos, pVal

Figure 3.2 Echo-Delimitation Algorithm Implemented in DE Key: preV – previous cM value, xL – limit of maximum echo length (~5.2 s), √ – store a value, × – erase a value. Other symbols are same as in Figure 3.1. 1 Algorithm of Echo Delimitation 79

Based on the requirements for least-squares fitting and Fourier transform (see below), the length of the selected echo must fall into a required range. The echo will be counted if it is longer than 0.6 s (192 sample points) for the RB samples or ~0.3 s (96 sample points) for the SB samples. The minimum analysable length of RB echoes is 0.8 s and for SB echoes is ~0.4 s. The maximum acceptable length is set at ~5.2 s. These length limits are based on considerations of increasing echo resolution and quality and decreasing number of false echoes, especially from rain droplets, as discussed in the following section.

2 Parameter Extraction

The IMR ran alternatively in RB- and SB-mode operations (see Chapter 2). Figure 3.3a shows the side view of beam geometry when the beam is rotating at a small offset angle to the zenith axis, and Figure 3.3b shows the plan view when the beam centre is at the azimuth ω, either an instant position under the RB-mode or one of the three parking positions under the SB-mode. IMR signal processing comprises analysis of digitised

Displacement (a) O (b) N Displacement C Displacement T (c)

Ω A E Orientation E r Orientation α T C Y Displacement β p' D r P' C L φ xτ X ∆ R O ω p y P δ L τ φ O p (d) P γ Ω γ

B B

Figure 3.3 IMR Beam Geometry: Side View (a) and Plan Views (b-d) The IMR rotates its beam centre C (with electric vector E) at a constant angular speed Ω about the zenith axis O at an offset angle δ (angular distance ∆) under the RB mode, or parks the beam electric vector E at one of three fixed azimuths at 60° interval under the SB mode. When an insect T crosses the IMR beam along a horizontal straight line L, it passes the point P of closest approach to the zenith at time τ, moving along the displacement direction γ with a body alignment β. Adapted from Riley et al. (1992a).

80 Chapter 3: Implementation of IMR Signal Processing

N DE.EXE yymmmdd Lyhhmmss.ddd? End

File to XMS N Y Ryhhmmss.ddd N Raining ? ? Demultiplexing Syhhmmss.ddd

N Y Range Echo available? available? N Y

Short Y Short echo echo? √@range -2

Selecting echo peak RB Echo Processing SB Echo Processing

Estimating F Estimating s trajectories trajectory minSize

P P

s Estimating Calculating γ, τ phase periodogram

P P Estimating Interpolating fFreq p, φ closest Wingbeat fPower approach Frequency P P P a0, a2, a4 Estimating Long echo Finding [hFreq β2, β4 size & shape √@range s ... harmonics hPower]

Figure 3.4 Flow Chart of IMR Signal Processing DE separates echoes from background noise and adjoining echoes and fits the top 6 dB of them to different models to extract target parameters, which are s – speed, γ – displacement direction, β – target orientation direction, a0 , a2 , a4 – RCS parameters, τ – time of closest approach to the zenith, p – angular distance of closest approach, φ– azimuthal angle of p, minSize –lower limit on RCS, fFreq – fundamental frequency (wingbeat), fPower – power of fundamental frequency, hFreq – harmonic frequency, and hPower – harmonic power. The wingbeat components can also be estimated from RB echoes if all parameters are well fitted to the models. P – Passes, F – Fails. 2 Parameter Extraction 81 signals from both RB- and SB-mode observations. DE processes a whole night’s observation data automatically according to its only command line parameter (or interactive input), yymmmdd (e.g. 98Apr30), which denotes both the date and the data directory. It reads in one data file (either Ryhhmmss.ddd or Syhhmmss.ddd) at a time, obtaining its name from the log file Lyhhmmss.ddd, separates it into corresponding gate ranges, and calculates periodic averages of signal strength for echo delimitation. The delimited short echoes are counted only, while the long ones are fully analysed to estimate the retrievable parameters. Figure 3.4 summarises the whole procedure and its details are described as follows.

2.1 Parameter Extraction from a Stationary-Beam Echo

The target parameters that can be estimated from an echo detected by an IMR operating in the SB mode are the speed, a lower limit for the RCS, and the wingbeat frequency and its modulation characters. Wingbeat frequency had been regarded as potentially valuable for target identification (e.g. Schaefer 1970), and its retrieval is the main motivation for operating on IMR in the SB mode. Although a basic SB-echo analysis method was present in the prototype program FIT, it has required significant development. A new implementation of the variable-length FFT and peak-fit algorithm was incorporated for fast and accurate frequency estimation. Windowing was introduced into the spectral analysis procedure to reduce frequency leakage (sidebands) and eliminate false frequency components. The characteristic form of the SB echo, its parameterisation, and the methods used in the program DE to extract these parameters are described with an example.

2.1.1 Signal Composition of Stationary-Beam Echo The power (W) reflected from a point target is given by the radar equation

pg22λ σ p = t 3.1 r 64π 34d

where pt is the transmitted power, g is the antenna off-axis gain, λ is the wavelength of the transmitter, d is the target range, and σ is the target’s RCS (Rinehart 1997). When an insect flies across the stationary IMR beam horizontally at a constant speed and with an unchanging body alignment, the antenna gain g from the IMR’s Gaussian beam is

82 Chapter 3: Implementation of IMR Signal Processing

1 2 gr()=− g0 exp( 2 kr) 3.2 where r is the angular distance between the target and the radar beam centre C (Figure

22 3.3), k = 8log2, and g0 is the on-axis gain π θ (θ is the half-power beam width). At any time t,

rt2222()=+ p ' st ( −τ ') 3.3 where τ ' is the time when the insect passes closest point P ' (at distance p ' ) to the beam centre C, and s is the insect’s speed (Figure 3.4c). Letting

pg22λ K = t 0 3.4 64π 3 i.e. the radar constant, and substituting Equations 3.2 and 3.3 into Equation 3.1, the power received is

K 2 ptr ()=−4 exp() krt ()σ () t d

K 2 =−+−expkp⎡⎤ '22 s() tτσ ' ( t ) 3.5 d 4 {}⎣⎦ K 22⎡⎤2 =−4 exp()kp ' exp −− ks() tτ 'σ ( t ) d ⎣⎦ where all terms before the second exponential function remain constant as the insect traverses the beam. The modulation of the RCS σ ()t will mainly be due to wingbeating assuming the insect does not yaw, which for insects is relatively minor. Converting the received power into the conventional logarithmic units of decibels relative to 1 mW (i.e. dBm), Equation 3.5 can be rewritten as

Pt( )=+ 10⎡⎤ 3 log K − 4log d − kp '22 log e − kst −τ '2 log e + logσε ( t ) + 3.6 r ⎣⎦10 10 10() 10 10 which shows that the signal intensity has the expected quadratic variation with time, the variation of the RCS term, assuming the wingbeat modulation is relatively small, will appear as an added ripple, and ε is the system noise.

2.1.2 Parameterisation of Stationary-Beam Echo For an SB Echo (Figure 3.5a, which shows the timeseries of signal intensity in parabolic shape with the modulation mainly from the insect’s wingbeating), the target speed and 2 Parameter Extraction 83 the lower limit on RCS are extracted in the time domain, while the wingbeat frequency and harmonics are estimated in the frequency domain.

650-700 m, 32o, 00:18:39.1~41.5 h, 27 Feb 2000, Bourke

-65

-70 2nd-order Polynomial Fit R2=0.932415, SD=0.913748

-75 (b) Modulation (mV) 0.15

0.00 -80 (c) -0.15

Signal Power (dBm) Power Signal (a) Hamming Windowing -85 0.15

(d) 0.00

-90 -0.15

0.0 0.4 0.8 1.2 1.6 2.0 Time (s) Figure 3.5 Example of Analysis of an SB Echo (1) (a) an SB echo about 2.4 s long was detected at Bourke from an altitude of 650-700 m, at 00:18:39.1 h on 27 Feb 2000, when the IMR beam E-vector was pointing to 32° azimuth. (b) A region of 1.536 s long (480 sample points), the peak, is selected from the echo for analysis. It is fitted with a second-order polynomial by weighted least-squares fitting; both the selected region and the fit curve are plotted 2.5 dB down from the original signal. (c) The trajectory component is subtracted from the signal, leaving a residual due mainly to the modulation from wingbeats. (d) The Hamming windowing function is applied to the modulation residual before carrying out the fast Fourier transform.

Estimating the speed As the echo strength is affected by the quadratic term of target traversing time, its parabolic character can be used to estimate the target speed. Because the time τ ' and the closest approach distance p ' are both unknown in Equation 3.6 but remain unchanged when the insect target traverses the radar beam, the measured signal intensity can be fitted to a general second-order polynomial (expanded Equation 3.6)

2 Ptraaa()=++ a bt ct 3.7

2 to the sampling time, where cksea =−10 log10 . The fitting is done by a weighted least- squares method with the usual weighting by the inverse square of measurement error

84 Chapter 3: Implementation of IMR Signal Processing

(Golub & Reinsch 1971; Gentle 1998). With the coefficient ca from the parabola with downward opening, i.e. ca is negative, the target speed s (in units of beam widths per second, BW/s) and its uncertainty σ s are estimated

⎧ 1 −c ⎪s = a ⎪ 45log210 ⎨ 3.8 1 σ ⎪σ = s c ⎪ s ⎩ 2 ca

where σ c , the standard deviation of the quadratic term ca , is an output of the least- squares fit. The speed error is calculated for evaluating the precision of the speed estimate. The linear speed estimate (unit meter per second, m/s) is then obtained by multiplying its value in beam widths per second by 2tan(/2)d θ .

The DE procedure adopts a top-down approach to select the signal part for processing. It first truncates the signal at both ends at the points where its periodic averages are about 6 dB down from the peak value. A truncation of a half length of the averaging period is made at any end that does not reach the level of 6 dB down, as this indicates that this end adjoins another echo. This second truncation reduces the chance of contamination by the echo from a preceding or following target. If the selected part is shorter than 0.1536 s (48 sample points), it will be counted but not analysed in any way, while very long signals are cut down to a maximum length of 3.2768 s (1024 points). The selected part is then fitted to Equation 3.7 and the fitted parabola is shown as Figure

3.5b. If the parabola proves to be inverted (ca > 0 ), or the time of closest approach to the beam centre,

b τ ' =− a 3.9 2ca is not in the selected signal range, the parameters are rejected and the echo is not processed further.

Estimating the lower limit for the RCS Although the RCS cannot be measured accurately from the SB signal, because the closest approach distance p ' to the beam centre cannot be determined, a lower limit

2 on the RCS can be estimated, by removing kp'log10 e term from Equation 3.6, as 2 Parameter Extraction 85

1 log10σ min= 10 PKd max−− 3 log 10 + 4log 10 3.10

where Pmax is the estimated echo power at the signal peak (t =τ '). Pmax can be calculated from the fitted second-order polynomial coefficients

2 ba Pamax =−a 3.11 4ca

2 2 The σ min has the unit of m which is converted to cm in the DE.

Estimating the wingbeat frequency Wingbeats appear as a ripple component along the signal. Thus the wingbeat frequency can be estimated from the fluctuations in the residual of the measured signal time series obtained by subtracting the parabolic trajectory component of Equation 3.6 (Figure 3.5b, c). The wingbeat modulations are then Fourier transformed into the frequency domain. The wingbeat fundamental frequency and its harmonics are estimated from the periodogram (i.e. the power spectrum). This processing is done in four steps.

Firstly, the trajectory components are subtracted from the original signal and the residues are converted from power (dBm) back to echo ‘voltage’ (arbitrary units) using the formula v =10P /20 . This is the inverse of the square-law detection and logarithmic transform done to the echo in the receiver. Figure 3.5c shows the residual modulations.

Secondly, the residual signal is multiplied by a “Hamming window” weighting function (Smith & Smith 1995) (Figure 3.5d, Table 3-1), in order to reduce frequency leakage, which otherwise will arise from the discontinuities at the ends of a finite-size sample (Breitenbach 1999). This also reduces contamination due to a poor polynomial fit or signals from nearby targets, as these tend to be worse further from the peak. Zeroes are added to the end of the signal (“zero padding”) to bring the signal length up to an exact power of two, as required for the fast recursive decomposition of the new FFT procedure (Press et al. 1998).

Thirdly, the signal is transformed from a time series of ‘voltages’ to a sequence of complex frequency components by FFT

N −1 ⎛⎞2πijk Ccwkjj==−∑ exp⎜⎟ k 0,..., N 1 3.12 j=0 ⎝⎠N

86 Chapter 3: Implementation of IMR Signal Processing

where c j is the input signal, wj is the windowing function, and N is the zero-padded signal length, which ranges from 64 ( 26 ) to 1024 ( 210 ) with the increase of power of two. The power spectrum is constructed from the real and the imaginary parts of the FFT output, as

⎧ 1 2 pf()= C ⎪ 00W ⎪ ss ⎪ 1 22 ⎛⎞N pf( )=+⎡⎤ C C k =− 1,..., 1 3.13 ⎨ kkNk⎣⎦− ⎜⎟ ⎪ Wss ⎝⎠2 ⎪ 1 2 ⎪ pf()NN/2= C /2 ⎩ Wss where k are the frequencies between 0 and the Nyquist frequency, which is half of the sampling frequency fS (312.5 Hz) (Press et al. 1998), and the normalisation factor is

Ns −1 2 WNwnss = ∑ () 3.14 n=0

where Ns is the real signal length (before zero-padding). Figure 3.6 shows the spectrum from the echo of Figure 3.5. The frequency resolution ∆ fS= f N , ranges from 0.31 to 4.88 Hz for the longest (1024) and shortest (64) FFT lengths respectively.

H o f 650-700 m, 32 , 00:18:39.1~41.5 h, 27 Feb 2000, Bourke

H H h3 h2

H h4 H h5 Modulation RelativePower 0 20 40 60 80 100 120 140 Wingbeat Frequency (Hz) Figure 3.6 Example of Analysis of an SB Echo (2) The Periodogram, from the SB echo shown in Figure 3.5, is used to locate the fundamental and harmonics, which are marked as Hf , Hh2 ,...,H h5 . The average spectral power is shown by a horizontal blue line.

Finally, the power in each spectrum bin and the energy of the spectral peak are used to identify frequency components. A peak-finding procedure searches through the spectrum for the peaks from the fundamental frequency and its harmonic components and calculates their peak position, i.e. the frequency. For identification as a fundamental frequency, a peak is required to include the strongest bin in the spectrum, which should 2 Parameter Extraction 87 be at least 8× stronger than the average from the whole spectrum, have an energy content that is at least 10% of the echo’s total spectral energy, and have a frequency higher than 14 Hz, this being the lowest wingbeat frequency reported for insects (Schaefer 1976; Riley 1979). The strongest power spectrum bin, which is normally from the wingbeat frequency, is to be located. The peak position is interpolated by a quadratic estimator (Donadio 1999)

pf()kk+−11− pf () ⎛⎞N kkP =+ k =0,...⎜⎟ − 1 3.15 ⎝⎠2 22()pf (kk )−− pf (−+11 ) pf ( k ) where k is the bin number of the power peak. The fundamental frequency is then calculated as fk=∆pf. Once the fundamental frequency is located, its harmonics are searched for at their expected frequencies, with a tolerance of ±2 bins. If a peak has the required frequency and its power is greater than 1.5× the spectral average, it will be identified as a harmonic.

Figure 3.7 Screen Capture of an SB Echo Analysis by the Program DE Analysis result of the SB echo as shown in Figure 3.5 and Figure 3.6 when DE is in the interactive mode.

88 Chapter 3: Implementation of IMR Signal Processing

Figure 3.7 shows the screen capture from the program DE for this SB echo illustrated in Figure 3.5 and Figure 3.6. It shows the delimited echo, the trimmed part for parameter estimation which has the fitted trajectory parabola on, the residual signal due to wingbeating, and the power spectrum with identified fundamental wingbeat frequency and its harmonics. The spectrum average is shown as a horizontal line. The target speed and the lower limit on the RCS are also printed on the screen, along with the coefficient and the standard error of the least-squares fit.

2.1.3 Implementation Optimisation of Spectral Analysis Some hardware and software incorporated to improve the performance of the signal analysis, and to speed the computation, are now described.

Antialiasing Aliasing is the appearance of false frequency components generated by frequencies above the Nyquist frequency. To prevent aliasing, a low-pass hardware filter is incorporated into the IMR’s signal-processing system (see Chapter 2). It has a cut-off frequency of 128 Hz, which is sufficient for the relatively low wingbeat frequencies from most noctuids and acridoids (Riley & Reynolds 1979) to pass, although the higher frequencies occurring mainly in Diptera and Hymenoptera (Hyatt & Maughan 1994) will be lost. The IMR digitises signals at 312.5 Hz in the SB mode of operation, and as required the resulting maximum spectral frequency, the Nyquist frequency (156.25 Hz), is higher than the cut-off of the antialiasing filter.

Reducing Frequency leakage Frequency leakage, which arises from the sharp discontinuities at the sample ends if the signal is not an exact number of cycles of the frequency repetition, is a common problem in spectral analyses that use a Fourier transform. It causes at least two problems: the combination of noise into signal energy, and the distortion or obscuration of weaker frequency components close to a strong frequency through the appearance of sidelobe peaks (Spyers-Ashby et al. 1998). Spectral leakage will affect the estimation accuracy of the amplitude and position of the frequency component.

A number of methods are available for reducing the frequency leakage (Sedlacek & Titera 1998; Breitenbach 1999), including splitting the data into shorter segments and averaging the resulting series of periodograms, and, if the signal period is 2 Parameter Extraction 89 known, coherent sampling or resampling of the data by shifting the sampling frame in the time domain (Lu et al. 1998; Sherlock 1999). However, these methods are not practicable for the use in the processing of IMR data. There is no way to adjust the sampling time so that an exact number of cycles of the frequency repetition will be taken, because the frequency is initially unknown. Resampling and other computation- consuming methods are not suitable either, as processing speed is important in IMR operations while the estimates of wingbeat frequency do not need to be highly accurate. Windowing in the time domain prior to the Fourier transform and then interpolating the frequency peak in the frequency domain (Andria et al. 1993) have thus been adopted in the processing of IMR echoes. This method scales the signal smoothly to nearly zero at the sample ends, so there is no energy to be spread, and therefore there are nearly no discontinuities.

Table 3-1 Windows and Figure of Merit Window Highest Sidelobe Frequency Coherent Equivalent 3-dB function sidelobe fall-off straddle gain noise mainlobe level ratio loss (dB) bandwidth bandwidth (dB) (dB/Oct) (bins) (bins) -13 -6 3.92 1.00 1.00 0.89 Rectangular wn( )== 1 n 0,..., N − 1

-32 -18 1.42 0.50 1.50 1.44 Hanning wn( )=− 0.5 0.5cos(2π n / N ) n = 0,..., N − 1

-43 -6 1.78 0.54 1.36 1.30 Hamming wn( )=− 0.54 0.46cos(2π n / N ) n = 0,..., N − 1

-58 -18 1.10 0.42 1.73 1.68 Blackman wn( )=− 0.42 0.50cos(2π nN / ) + 0.08cos(4π nN / ) n = 0,..., N − 1

Four- -92 -6 0.83 0.36 2.00 1.90 sample wn( )=− 0.35875 0.48829cos(2π nN / ) + 0.14128cos(4π nN / ) Blackman- −=−0.01168cos(6π nN / ) n 0,..., N 1 Harris Reproduced from Smith and Smith (1995).

A variety of windowing functions has been developed for this purpose (Smith & Smith 1995). Selection of the most suitable one for a particular application is based on the effects it has on frequency resolution (the width of the mainlobe), spectral leakage (the height of the highest sidelobe) and amplitude accuracy (the flatness of the mainlobe). Since the frequency is of most concern here, the narrower the 3-dB

90 Chapter 3: Implementation of IMR Signal Processing bandwidth of the mainlobe, the better. Estimating the absolute amplitude of the frequency is not the first priority in this case. With a rectangular window (i.e. no weighting), sidelobes close to the main frequencies were often seen in the IMR signals. It was first thought that they might be the result of the insect “breathing”, as suggested by Schaefer (1970) in regard to his observations of Desert Locusts, but it turned out that they are due to spectral leakage, and that applying any window function (other than the rectangular) eliminates them completely. The four-sample Blackman-Harris window was used initially (Hyatt & Maughan 1994; Drake et al. 2002b). Later it was realised that this window provides much more sidelobe suppression than is needed, since the IMR signal noise is not severe and the improved echo-delimitation algorithm has greatly reduced the interference from nearby targets. Table 3-1 lists the characteristics of five commonly used window functions and Figure 3.8 shows their forms.

1.0 Rectangular

0.8

Hanning Hamming 0.6

Blackman

Amplitude 0.4

Four-sample 0.2 Blackman-Harris

0.0

0 64 128 192 256 Sample Length Figure 3.8 Characteristics of Often-Used Windows Five weighting functions, which are often used in spectral analysis, are drawn for a signal 256 points long. See Table 3-1 for the equations.

The Hamming window, which delivers a good performance compromise between frequency resolution and suppression of spectral leakage (Smith & Smith 1995), is now used. Figure 3.9 presents the effects of various windows on the spectral- 2 Parameter Extraction 91 density estimation of an echo. It is obvious that any of these windows, except the rectangular window, is sufficient to reduce spectral leakage and noise to an acceptable level. Although the Blackman and the Blackman-Harris windows have the best noise suppression, their broad mainlobes produce less accurate frequency positions. The Hamming window, on the other hand, has the narrowest frequency resolution, so a more precise frequency estimate can be expected.

31.54Hz 350-400m, 21:18:34.4512~35.3728h, E-vector 32°, 28 Feb 2000, Bourke -Harris 4s Blackman 63.02Hz 94.50Hz

31.60Hz

94.50Hz Blackman 62.99Hz

31.60Hz

94.52Hz Hamming 62.96Hz

31.65Hz

Hanning 94.51Hz 62.97Hz

31.29Hz

94.55Hz 62.93Hz Rectangular 0 20 40 60 80 100 120 140 Frequency Figure 3.9 Comparison of Window Functions on Periodogram Estimation Five weighting functions are applied to a 0.9216-s SB echo, which was detected at Bourke at an altitude of 350-400 m, at 21:18:34.5 h on 28 Feb 2000, when the IMR beam E-vector pointed to 32° azimuth. The average signal energy is marked as a horizontal line (blue, above the background level). The fundamental frequency and harmonics have been interpolated by the quadratic estimator of Equation 3.15. Some noise peaks and spectral leakages are marked with red circles.

The increase in width of the mainlobe that results from windowing may also prevent close frequency components from being resolved. For example, the frequency that appears on the right side of the fundamental peak at 31.29 Hz (decimal frequency to show the differences of precise peak position only, Figure 3.9) where the rectangular window is used, may not be a frequency leakage, while that on the left side almost certainly is. However, both side peaks disappear totally under the strong sidelobe suppression of the Blackman-Harris four-sample window. At any rate, this is unlikely to be a severe problem in the estimation of wingbeat frequency, in which usually only a

92 Chapter 3: Implementation of IMR Signal Processing single main frequency will be present and biological reasons will make accuracy to better than 1 Hz of no use. Whether the sidelobes from breathing are retrievable is yet to be investigated.

Frequency interpolation The accuracy of frequency estimation is affected by two sources of error: the defect of windowed Fourier transform (Harris 1978), and scalloping loss (straddle loss, or picket-fence effect), i.e. attenuation due to frequencies that do not coincide with FFT bin centres (Smith & Smith 1995). The former cannot be eliminated, as the physical resolution of the Fourier transform directly depends on the number of data samples, while the latter can be reduced by interpolation in the frequency domain (Goto 2000; Zhang et al. 2001). Zero padding is often used to reduce scalloping loss and increase computational resolution and does not provide any additional information, while it does increase the computational load. A more effective approach is to use peak interpolation to estimate the frequency (Kootsookos 1997; Jacobsen 2001).

Quinn’s second estimator of peak interpolation (Quinn 1994) is the best method available, as it has very high accuracy under almost any signal condition. However, the quadratic estimator of Equation 3.15 is very competitive in frequency accuracy over a very wide range of signal-to-noise ratios (Donadio 1999), and is easy to implement and requires less computation. It has therefore been adopted for the wingbeat retrievals, along with the minimum amount of zero padding in the time domain. The use of a specific interpolation algorithm for a particular windowing function to achieve very high frequency accuracy, as suggested by Goto (2000) and Zhang et al (2001), has not been considered due to its complexity and the limited accuracy requirement for this application.

With these improvements and the rewritten LSF and FFT codes, DE provides fast, stable and accurate parameter estimation for the signals from IMR SB-mode observations.

2.2 Parameter Extraction from a Rotary-Beam Echo

With a rotating offset radar beam, the target’s position in relation to the radar beam axis can be determined and its trajectory estimated. This in turn allows the target’s size and 2 Parameter Extraction 93 shape to be measured. Six parameters – speed, displacement direction, body alignment direction, and three body size and shape parameters – can be extracted from a rotary- beam echo (Bent 1984; Riley et al. 1992a; Smith et al. 1993; Harman & Drake 2004). In addition, the wingbeat frequency can be estimated from a good-quality echo (Wang & Drake 2004). The RB echo pattern, extraction of target information from it, and implementation of the computation are described in this section.

2.2.1 Signal Composition of Rotary-Beam Echo Assuming an insect T approaches its closest approach P to the zenith axis O at time τ while traverses the rotating IMR offset beam (Figure 3.4d), within a rectangular Cartesian coordinate system, its position at time τ related to the origin O can be described as

⎧Xpτ = sinφ ⎨ 3.16 ⎩Ypτ = cosφ where p is the distance of closest approach P to the zenith axis O, and φ is the azimuthal angle of p from the north (Y axis). Generally, at any time t, the position of the insect T can be described as

⎧Xpt =+−sinφ st (τγ )sin ⎨ 3.17 ⎩Ypt =+−cosφ st (τγ )cos where s is the insect’s constant speed (in BW/s), and γ is the direction of the insect’s displacement. The angle γ has the relationship φ = γ ±° 90 to φ (Figure 3.4b), indicating whether the insects passes to the right (+) or the left (-) of the zenith axis O. However, to simplify the computation, both of them are used and estimated independently. The instantaneous beam centre at any time t is at

⎧X ct=∆sinω ⎨ 3.18 ⎩Yct=∆cosω

where ∆ is the angular distance of the beam offset δ, ωt = Ω+t ξ , Ω is the angular speed of the beam conical scan, and ξ is the initial azimuthal angle of the polarisation (the electric vector E) if it is not coincident with the north. The squared angular distance of the insect to the beam centre is

94 Chapter 3: Implementation of IMR Signal Processing

222 rXXYY=−()()tc +− tc 3.19

Substituting Equations 3.17 and 3.18 and incorporating some trigonometric relations, Equation 3.19 can be represented as

2222 2 rpstp=∆ + +()2cos()()cos() −τ − ∆[ ωφtt − + st − τ ωγ − ] 3.20

As the IMR beam is linearly polarised and continuously rotated, without consideration of the wingbeating modulation, the radar cross-section σ is a function of the instantaneous angle α =−ωβt between the polarisation plane ω and the insect’s longitudinal body axis β (Aldhous 1989; Smith et al. 1993)

σ (ααα )= aa++ cos2 a cos4 02 4 3.21 =+aa02cos2(ωtt −βωβ 2 ) + a 4 cos4( − 4 )

where β2 in the range of 0 – 180° and β4 in the range of 0 – 90°; both denote to the orientation direction. The coefficients of , a and a are real positive constants a0 2 4 determined by

⎧a =++3 ()σ σσσµ1 cos ⎪ 0 84xx yy xx yy ⎪ 1 ⎨a2 =−2 ()σσxx yy 3.22 ⎪ a =+−11()σ σσσµ cos ⎩⎪ 4 84xx yy xx yy

⎛⎞σ 0 where µ is the phase angle in the radar scattering matrix S = ⎜⎟xx , σ is ⎜⎟0 eiµ σ xx ⎝⎠yy the maximum RCS value (usually when the polarisation plane is parallel to the longitudinal axis of the insect’s body), and σ yy is the RCS value when the polarisation plane is perpendicular to that of σ xx (Smith et al. 1993). Substituting Equations 3.4, 3.20 and 3.21 into the radar equation Equation 3.1, the power received is

K 2 ptr ()=−4 exp() kr σ d

K 222 2 =−∆++−−∆−+−−4 expkpst{} (τωφτωγ ) 2[] p cos(tt ) st ( )cos( ) 3.23 d ()

[]aa02+−+−cos2(ωβtt 2 ) a 4 cos4( ωβ 4 ) 2 Parameter Extraction 95

Converting the received power into logarithmic units (dBm), Equation 3.23 can be rewritten as

K Ptr ( )=+ 30 10log10 ( ) + 10log 10[] a 0 + a 2 cos 2(ωβtt − 2 ) + a 4 cos 4( ωβ − 4 ) − d 4 3.24 222 2 10ke log10 ( ){}∆+ pst + ( −τ ) −∆ 2[] p cos(ωφtt − ) + st ( − τ )cos( ωγ − ) + ε

with the system noise ε. In Equations 3.23 and 3.24, K, k, ∆, and ωt are known constants; s, γ, β2 (β4 ), a0 , a2 and a4 are six parameters related to the insect’s flight and its size and shape; while τ, p and φ are three intermediate parameters that also have to be determined as part of the calculation.

2.2.2 Parameterisation of Rotary-Beam Echo The estimation of target parameters from an RB echo is all done in the time domain using polynomial and trigonometrical polynomial least-squares fits. This method, developed primarily by I. T. Harman, was adopted in preference to the more complex frequency-domain method of Smith et al (1993). The implementation described here is that used in DE and differs in some details from that in Harman and Drake (2004).

The time series of signal intensity produced when an overflying insect traverses the conically scanning IMR beam is considerably more complicated than that produced in the SB mode, as three kinds of modulation are superimposed on the basic parabolic form. The deepest modulation usually arises from the variation of radar cross-section with beam polarisation and is a result of the elongated cylindrical shape of the insect’s body (Riley et al. 1992a; Smith et al. 1993). Except in the case of the largest species, maximum power is reflected back when the insect’s longitudinal body axis is parallel to the electric vector of the radar beam (Riley 1973; Schaefer 1976; Riley 1978; Drake 1981a). Therefore, longer and thinner insects usually produce deeper and narrower polarisation modulations than shorter and broader ones (Riley 1978). The second type of modulation arises from the beam nutation, i.e. the beam rotates at an small offset angle δ to the zenith axis O, and depends on the angular speed of beam rotation, the distance of closest approach p and the speed of the overflying insect (Drake et al. 1998). The weakest modulations are usually caused by the insect’s wingbeating. Figure 3.10 shows a typical RB echo signal together with its simulation (without wingbeat modulation) calculated from the set of extracted parameters.

96 Chapter 3: Implementation of IMR Signal Processing

Circular Average -60 (a) T ⎯⎯ Azimuth 0° ⊥ ⎯⎯ Azimuth 180°

-70

-80

-90 R echo, 550-600 m, 00:31:03.6~07.0 h, 27 Jun 2000, Bourke Time (s) (b) Real Signal Simulated Signal -60

Signal Power (dBm) Power Signal R , SD of least-squares fit -70 1) Trajectory Fit: 0.965273, 0.056796 2) Closest Approach Fit: -80 0.982423, 0.717735 3) Orientation Fit: -90 0.961207, 2.094965e-07

0.0 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2 Figure 3.10 Example of a RB Signal (a) and the Signal Reconstructed from Estimated Parameters (b) A 3.4-s RB echo was detected in the altitude range 550-600 m at 00:31:03.6 h on 27 Jun 2000, by the IMR at Bourke. A 1.8-s selection from this signal, centred at its peak, was used to extract target parameters, which were then used to simulate the original signal for evaluating the goodness-of-fit. The azimuthal angles of beam polarisation are marked when the conical scan passed through north and south. The power averages from each conical-scan cycle are plotted as a stepped line, which illustrates a parabolic shape and indicates the insect’s approach towards and departure from the beam. Target parameters: speed = 7.4 ± 0.2 m/s, displacement direction = 283°, orientation = 128° / 308°, a0 = 1.62 2 2 2 cm , a2 = 1.36 cm , a4 = 0.43 cm . Intermediate parameters: p = 0.43 beam widths (0.475°), τ = 0.89 s (from the beginning of selected part), φ = 189°≈283-90° which indicates that the insect passed to the left of the zenith axis.

Estimating the speed The target speed from an insect traversing the IMR’s rotating beam can also be estimated using the parabolic fit as for an SB echo (see page 83). The time series of signal power (Figure 3.11a) is truncated at the points when it is 6 dB down from the top according to the values of the periodic averages (Figure 3.11b). A further half length of the average cycle will also be trimmed if any end does not reach the 6 dB down from the echo peak. The selected signal part is sifted into 64 vectors, one for each azimuthal position of the rotating beam at which sampling occurs. The signal intensities sampled at the same azimuthal position in successive cycles are dependent on the angle α of the polarisation plane to the insect’s body axis, and the angular distance r of the insect from 2 Parameter Extraction 97 the instantaneous beam centre C, with only the latter term changing (provided that insect does not yaw) because all the cosine terms in Equation 3.24 are constant (Figure

3.4). The time-series of signal intensity Pti ( ) (dBm) at each of the 64 azimuthal positions are fitted by a quadratic model, as a parabolic shape is expected just as with an SB echo (Figure 3.11c),

2 Ptiiii( )=++ a bt ct ( i = 0,1,2,...,63) 3.25

Weighting is again by the inverse square of the error on the power measurement.

Therefore, at each sampling azimuthal position, the target speed si (BW/s) and its uncertainty σ are estimated by si

⎧ 1 −ci ⎪si = ⎪ 45log210 ⎨ (i = 0,...,63) 3.26 1 σ ⎪σ = s ci ⎪ sii ⎩ 2 ci

-60 T ⎯⎯ Azimuth 0° (a) ⊥ ⎯⎯ Azimuth 180° -70

-80

-90 700-750 m, 00:31:45.6-49.2 h, 27 Feb 2000, Bourke -60 (b) -70

-80 Secondary Selection o -90 Azimuth 353 Signal Power (dBm) -60 (c) -70

Azimuth 353o -80 2nd-order Polynominal Fit R2=0.99597, SD=0.27796 -90

0.0 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.6 Time (s) Figure 3.11 Example of Parameter Estimation from an RB echo (1) This figure shows a 3.6-s RB echo (a) detected by the IMR at Bourke from an altitude of 700-750 m, at 00:31:45.6 h on 27 Feb 2000. The 2-s selection (b), on which the samples taken at 353° azimuth are marked and fitted to a second-order polynomial (c), made from the peak region of the echo is used for parameter estimation.

98 Chapter 3: Implementation of IMR Signal Processing just as the same procedure for an SB echo (Equation 3.8). Any inverted parabola will be discarded and not be used in the further analysis. The target speed is then estimated as an arithmetic mean of the values obtained at all sampling azimuths

1 n ssinn==≤∑ i ( 0,..., , 63) 3.27 n +1 0

Estimating the flight trajectory

For each of the 64 azimuthal beam positions ωi , a version of Equation 3.24 can be written as

K Pt( )=+ 30 10log ( ) + 10log (σ ) − 10 k log ( e )⎡ ∆+−∆22 p 2 p cos(ωφ − )⎤ ii104 10 10 ⎣ i⎦ d 3.28 ⎡⎤22 −−−∆−−=10kestst log10 ( )⎣⎦ (ττωγ ) 2 ( )cos(i ) ( i 0,...,63)

from which the quadratic dependence on time t of the power Pi value for the particular angle ωi is evident. This simple time dependence arises as there is only a small variation of RCS from wingbeating and ∆ , p, ωi − β2 , ωi − β4 , ωi −φ , and ωi −γ are all constant for a particular trajectory and the beam azimuthal direction. Therefore, when

dP i =−−∆−==2st2 (τωγ ) 2 s cos( ) 0 ( i 0,...,63) 3.29 dt i

Pti () reaches its maximum, i.e. the maximum signal intensity is received at

∆ ⎧i = 0,...,63 ττ=+cos( ωγ − ) 3.30 ii⎨ π 63π s ⎩ωξξi =+,32 ,..., ξ + 32

This can be expanded as

∆∆⎧i = 0,...,63 ττ=+cos γ cos ω + sin γ sin ω 3.31 iii⎨ π 63π ss⎩ωξξi =+,32 ,..., ξ + 32

where the time τ i of the closest approach to the instantaneous beam centre at ωi can be calculated from ai , bi and ci , the fitted second-order polynomial coefficients from Equation 3.25, 2 Parameter Extraction 99

bi τ i =−(i = 0,1,2,...,63) 3.32 2ci

Therefore, Equation 3.31 can be fitted by a 3-parameter trigonometric polynomial,

π 63π τ ibb=+abcosωωωξξξ ib + c sin i ( i = , +32 ,..., + 32 ) 3.33

where cosωi and sinωi are the independent variables, and weighting is also by the inverse squared error on τ i (Figure 3.12a). From a comparison of Equations 3.31 and ∆ ∆ 3.33, a ≡τ , b ≡ cosγ and c ≡ sinγ , the trajectory parameters can be estimated b b s c s from the regression coefficients ab , bb and cb of Equation 3.33

⎧τ = ab ⎪ c ⎪γ = arctan(b ) ⎪ bb ⎪ ⎪ ∆ ⎨s = 3.34 ⎪ bc22+ ⎪ bb ⎪ 22 σ bc+σ ⎪σ s = s 22 ⎩⎪ bcbb+ where the time τ of closest approach to the zenith axis is in seconds, the displacement direction γ is in radians (range 0 – 2π), and the target speed s and its standard error σ s are in BW/s and converted to m/s.

Comparing target speeds from Equations 3.27 and 3.34, the measure of the IMR beam offset ∆ can be verified.

Estimating the closest approach As each pair of diametrically opposite azimuthal sampling positions have the same polarisation, the RCS measured should be the same, i.e. σ ()ωσωπii=+ ( ). Taking the differences of the power at these two points will eliminate the RCS terms from Equation 3.28. The power ratio from those diametrically opposite positions is, from Equation 3.28,

PPii−=+3220 k log 10 ( ep ) ∆ [cos(ω i −−−+φωφπ ) cos( i )] ⎧i = 0," ,31 ⎪ =∆−40kep log10 ( ) cos(ωφi ) ⎨ 31π 3.35 ⎪ωξi =+,," ξ =∆40kep log10 ( ) (cosφω cosii + sin φω sin ) ⎩ 32

100 Chapter 3: Implementation of IMR Signal Processing where the power for that azimuthal position can be estimated for t =τ by Equation 3.25 with the fitted coefficients, i.e.

2 Pabciiiii(τττ )=+ + ( = 0,1,...,63) 3.36

At the point of closest approach to the zenith (when t =τ ), the signal reaches it peak and therefore the effect from noise and interference of adjoining echoes are minimum. A 2-parameter trigonometric polynomial

31π Ra()ωωωωξξ=+ cos b sin ( =+ ," , ) 3.37 pcicii 32

is then fitted, where, by comparison with Equation 3.35, akepc ≡ 40 log10 ( )∆ cosφ , bkc ≡∆40 log10 ( ep ) sinφ . The inverse sum of the squares of the standard deviations of

Trajectory Fit 1.6 (a) R2 = 0.984936 SD= 0.040632 1.2

Phase (s) 0.8 700-750 m, 00:31:45.6-49.2 h, 27 Feb 2000, Bourke 0.4 4 Closest Approach Fit 2 (b) 2 R = 0.978427 SD= 0.309491 0

-2 Power Diff (dBm) Power Diff -4 -64 Orientation Fit (c) 2 -66 R = 0.995630 SD= 1.780714e-08 -68 -70

Power (dBm) -72 -74 90 180 270 360 Azimuth Angle (deg)

Figure 3.12 Example of Parameter Estimation from an RB echo (2) Continued from Figure 3.11. The times of closest approach to the instantaneous beam centres from 64 azimuthal angles and the standard deviations from the parabolic fits are used to fit a 3-parameter trigonometric polynomial, to estimate the time τ when the insect passed the closest approach to the zenith, the displacement direction γ and the speed s in (a). These times of closest approach to the beam centres are also used to estimate the maximum signal intensities, which are used to fit a 2-parameter trigonometric polynomial to the intensity differences at opposite polarisation angles, to estimate the distance p of closest approach to the zenith and its azimuthal angle φ in (b). The estimated RCS at each sampling azimuthal angle are used to fit a 5-parameter trigonometric polynomial, to estimate the orientation β, and three size and shape parameters a0 , a2 and a4 in (c). 2 Parameter Extraction 101

the Pi values calculated from Equation 3.25 for the two opposite beam positions are used as the weight. Hence, the angular distance p of closest approach to the zenith, and its azimuthal angle φ (clockwise from north) are estimated from the Equation 3.37 regression parameters as

⎧ ab22+ ⎪ p = cc ⎪ 40ke∆ log ( ) ⎨ 10 3.38 ⎪ bc ⎪φ = arctan( ) ⎩ ac where p is in BW (converted to radians in DE) and φ is in radians in the range of 0 – 2π (Figure 3.12b). φ can be used to crosscheck the estimate of the displacement direction γ and determine whether the target passes to the left or the right of the zenith.

Estimating the RCS The received power from each of the 64 sampling azimuthal angles at time t =τ can also be used to estimate the target orientation, body size and body shape parameters.

From Equation 3.21, the target RCS σ i at any sample azimuth can be represented as

σ ii= aa02+−+−cos 2(ωβ 2 ) a 4 cos 4( ωβ i 4 ) ⎧i = 0," ,63 ⎪ =+aa02cos 2βω 2 cos 2ii + a 2 sin 2 βω 2 sin 2 ⎨ 63π 3.39 ⎪ωξi =+,," ξ ++aa44cos 4βω cos 4ii 44 sin 4 βω sin 4 ⎩ 32

At the point of closest approach when time t =τ the target RCS can be estimated from Equation 3.23 as

pd()ττ44 pd () σωφ==iiexpkpp⎡⎤ ∆+−∆−=22 2 cos( ) ( i 0," ,63) 3.40 i −kr 2 {}⎣⎦i Ke i K

with the substitution of Equation 3.25 for pi (τ ) . This set of 64 measurements of the target RCS can be fitted by a 5-parameter trigonometric polynomial

63π σ (ωωωωωωξξiddidididii )=+ab cos 2 + c sin 2 + d cos 4 + e sin 4 ( = ,..., +32 ) 3.41

where, by comparison with Equation 3.39, aad ≡ 0 , bad ≡ 22cos 2β , cad ≡ 22sin 2β , dad ≡ 44cos 4β , and ead ≡ 44sin 4β . RCS parameters and orientation are then estimated as

102 Chapter 3: Implementation of IMR Signal Processing

⎧aa0 = d ⎪ 22 ⎪abc2 =+dd ⎪ ade=+22 3.42 ⎪ 4 dd ⎨ 1 cd ⎪β2 = arctan( ) ⎪ 2 bd ⎪ 1 ed ⎪β4 = arctan( ) ⎩⎪ 4 dd

2 2 where the RCS parameters a0 , a2 and a4 are in the unit of m and converted into cm and β2 and β4 are in radians (Figure 3.12c). Both β2 and β4 are estimated and output to the result file, although β2 is often used directly as the orientation direction.

Estimating wingbeat frequency Recent investigations have shown that when all these nine parameters are extracted successfully from an RB echo, and the wingbeat modulations are strong, the

Figure 3.13 Screen Capture of an RB Echo Analysis by the Program DE This is the result for the echo shown in Figure 3.11 and Figure 3.12. The target speed is estimated as 7.49 and 7.59 m/s by Equations 3.27 and 3.34 respectively. The RCS coefficients are a0 = 0.85 , a2 = 0.77 2 and a4 = 0.24 cm . The estimated displacement and orientation directions are plotted on the polar diagram, with the scaled p to the origin. The target shape is indicated roughly by the ellipse. 2 Parameter Extraction 103 wingbeat frequency can also be extracted (Wang & Drake 2004). Once parameter retrieval for an RB echo is completed, the time series of echo power can be reconstructed from Equation 3.24 using the estimated parameter values (Figure 3.10b). If this estimated power (in dBm) is then subtracted from the observed signal power, the residual (i.e. the ratio of real signal to simulated signal, see page 85) would be mainly due to the insect body distortion associated with wingbeating, apart from the system noise. Thus spectral analysis, and extraction of wingbeat frequency and other wingbeat characters, exactly as for an SB echo, should be possible. This was found to be the case and the DE algorithm was modified to incorporate wingbeat retrieval from RB echoes routinely.

Figure 3.13 shows the full analysis result of the RB echo, the processing of which has been illustrated step by step in Figure 3.11 and Figure 3.12. In this output from DE’s interactive mode, the least-squares fits and the periodogram are plotted for view inspection, along with the original echo and its simulation. The periodogram is estimated from the residual after subtracting the simulated signal and converting to a

Figure 3.14 Example of Wingbeat Frequency that cannot be Extracted from an RB Echo The wingbeat frequency may be at 30 Hz. However, it seems to be masked by the 6th harmonic of the conical scan, which is 5 Hz.

104 Chapter 3: Implementation of IMR Signal Processing

‘voltage’. The fundamental frequency is strong and clear at 33.0 Hz, with two harmonics at 65.9 and 98.9 Hz, respectively. This is a typical good-quality echo. If the six parameters are not estimated with high accuracy, the wingbeat frequency may be masked by noise and peaks of the conical scan rate and its harmonics. Figure 3.14 shows an example in which the wingbeat frequency is masked in this way and therefore cannot be determined.

2.2.3 Improvements of Computation Several improvements have been made to enhance the computational performance of DE for routine signal processing. A weighted least-squares algorithm using the Golub- Reinsch singular value decomposition (Golub & Reinsch 1971; Gentle 1998) has been implemented into the program DE, replacing a previous Givens rotations and singular value decomposition algorithm (Nash 1979). SVD remains the best method to solve most linear squares problems, of both full and deficient rank. However, the original least-squares fit code cannot handle ill-conditioned data very well, producing occasionally either overflow/underflow error from extreme values or endless iterative loops. The Golub-Reinsch algorithm has been chosen as it is almost a SVD standard. It has been extensively tested, very stable and efficient, and widely used in many software, including EISPACK, LINPACK, LAPACK and MATLAB. The least-squares fit function has been generalised to solve the 2-, 3- and 5-parameter trigonometric polynomials, as well as the second-order polynomial. This simplifies the code and increases the code readability. Computational failures, which occurred very frequently when the original FIT implementation was applied to routine processing of the IMR data, have been eliminated. Other code optimisations have also increased the program

2 stability. For example, to calculate the second-order polynomial f ()taatat=++01 2 as in Equation 3.25, a recursive sum of multiplications

2 ⎧yt0 ()= 1 ft()= ∑ aytii () ⎨ i=0 ⎩ytii()= ty−1

2 is adopted instead of the power function f ()tat= i , to avoid possible domain (i.e. ∑i=0 i not the required input argument value defined by the function, such as calculation of t 2 2 Parameter Extraction 105 for t = 0 ) or underflow errors when t → 0 (very small, exceeding the machine precision). The Euclidean distance function ab22+ has also been replaced by

2 ⎧ ⎛⎞b ⎪ aab1+>⎜⎟ ⎪ ⎝⎠a ⎪ 2 ⎨ ⎛⎞a ⎪ bbab1,0+ ⎜⎟ ≤≠ ⎪ ⎝⎠b ⎪ ⎩0otherwise to avoid any overflow and underflow errors from extreme values due to machine precision. In addition, directly using expanded memory for storing raw radar data speeds the processing, which reduces the frequency of disk access and the use of conventional memory (limited to 640 KB, of which only about 520 KB is available for application programs after the DOS system is loaded). All these implementations have proven necessary for the stable and fast data process.

3 Some Measures of IMR Performance

For successful implementation of the IMR data processing as a fully automatic process, it is necessary to ensure that the computational efficiency and reliability are satisfactory. The new algorithm for echo delimitation has been examined for its sensitivity and reliability, and some results are summarised here to demonstrate the data quality from DE routine processing, as well as the hardware accuracy.

3.1 Performance of Echo-Delimitation Algorithm

To determine whether the automated echo-delimitation algorithm can reliably and precisely identify target echoes from the radar signal, five nights, which represent conditions of insect activity at low, intermediate and high levels, as well as of weather in clear, light and heavy rains, were chosen from the IMR observations at Bourke to examine the echo count and signal selection. Figure 3.15 shows the typical signal patterns from these nights. In addition, four schemes which differ in their requirements

106 Chapter 3: Implementation of IMR Signal Processing

dBm -70 (a) -80

-90 122.0 122.4 122.8 123.2 123.6 124.0 124.4 124.8 125.2 -80 R echo, 550-600 m, 18:03:37 h, 23 Jun 2000, Bourke -85

0 20 40 60 80 100 120 140 s -75 dBm -80 (b) -85 -90 -70 118.0 118.5 119.0 119.5

-80 S Echo, 800-850 m, 01:23:07 h, 21 Jun 2000, Bourke

-90 116 120 124 128 132 136 140 144 148 152 s -70 dBm (c) -80 -90

90.8 91.2 91.6 92.0 92.4 92.8 93.2 -70 R Echo, 650-700 m, 23:08:57 h, 21 Jun 2000, Bourke

-80 -90 84 88 92 96 100 104 108 112 116 120 s -60 dBm (d) -70 -80

-60 94.4 94.6 94.8 95.0 95.2 95.4 95.6 95.8 96.0 96.2

-70 S echo, 350-400 m, 18:23:43 h, 19 Jun 2000, Bourke

-80

96 100 104 108 112 116 120 124 128 132 s -30 dBm (e) -35 -40 -45 Not a valid echo 63.2 64.0 64.8 65.6 66.4 67.2 -36 R echo, 200-250 m, 18:00:56 h, 02 May 2000, Bourke

-38

64 68 72 76 80 84 88 92 96 100 s

Figure 3.15 Signal Patterns from Different Nights of Insect Activity and Weather Time series of period-averaged signal (lower plot of each pair) and example echo for a selected signal (upper plot of each pair), for five different observing conditions: (a) light, (b) intermediate and (c) heavy flight activity during fine nights, and (d) light and (e) heavy rains, for two observing modes: rotary-beam (R echo, a, c and e) and stationary-beam (S echo, b and d). 3 Some Measures of IMR Performance 107 regarding echo strength and completeness have been used to evaluate the sensitivity of the algorithm. They are:

1) 6dB full parabola: peak intensity at least 6 dB above either the threshold or the level at the boundary with adjoining echoes on both sides; 2) 6dB half parabola: peak intensity at least 6 dB as in 1), but a fall of this size is required on one side only; 3) 3dB full parabola: as 1), but with peak intensity reduced to 3 dB; and 4) 3dB half parabola: as 2), but with peak intensity also reduced to 3 dB. The effectiveness of echo-delimitation was assessed by the comparisons of the total number of echoes identified, the number of echoes that are analysable (with all parameters extracted successfully but possibly with poor fit to the models) and the number of echoes that make good fits on the IMR signals of these five nights. The quality of the identified echo was assessed using the goodness-of-fit measures produced from the least-squares fits during parameter estimation. High correlations show that the measured signal is well described by the theoretical model, indicating that the estimated parameters are more accurate. An RB echo is considered of good quality if the 3 coefficients of determination from trajectory, closest approach and orientation fits are all greater than 0.8, and an SB echo if the coefficient of determination from the trajectory fit is also greater than 0.8 and the fundamental frequency is identified.

3.1.1 Algorithm sensitivity The number of identified echoes from the same radar signal under each of the four different echo-selection schemes is a direct indicator of the sensitivity of the echo- delimitation algorithm. Figure 3.16 shows the results for the five selected nights; the echo counts are shown as ratios relative to those for echo-selection criterion 1), which is the most demanding one on signal condition. The numbers of good-quality, analysable, and total echoes identified by each of the four echo-delimitation criteria, for the whole of each of the five nights, are also given. Generally, the echo numbers increase as the echo-selection criteria become less demanding. The count increase is slight during the fine nights (a-c), less than 2-fold on the total counts, but large on the rainy nights (d, e), up to nearly 9-fold. In fine weather, the increases from criterion 2) to 3) are not very high in comparison with those from criterion 1) to 2) and 3) to 4); on rainy nights, however, there are increases at each stage. This indicates, as the echo-selection criteria

108 Chapter 3: Implementation of IMR Signal Processing

6dB 6dB 3dB 3dB

1.6 (a) RG 00Jun23 SG 1.4 fine RA SA 1.2 RT ST 1.0 68/353/361 73/423/479 73/431/460 76/465/556

(b) RG 1.6 00Jun20 SG fine RA 1.4 SA 1.2 RT ST 1.0 727/3203/3313 781/3977/4660 752/4332/4848 780/4662/5758 (c) 1.8 RG 00Jun21 SG 1.6 fine RA 1.4 SA RT 1.2 ST 1.0 1962/12460/12909 2150/16402/18423 2114/17942/19381 2238/20965/24087 (d) 8 RG 00Jun19 SG 6 rainy RA 4 SA RT 2 ST 76/310/336 86/442/648 77/451/1127 88/1427/2742 12 (e) RG 10 00May02 8 rainy SG RA 6 SA 4 RT 2 ST 135/705/731 157/995/1357 148/1441/1997 155/2874/5537 Figure 3.16 Variation of Echo Counts with Different Criteria for Echo Delimitation Count ratios of good echoes, of analysable echoes, and of total echoes (which includes short echoes and echoes from which all required parameters cannot be retrieved), are the sums of both R and S samples of 11 h observations. The rain-checking procedure was switched off when processing these data. Key: R – Rotary-beam echo, S – Stationary-beam echo, G – Good quality, A – Analysable, T – Total counts. 3 Some Measures of IMR Performance 109 become less demanding, most of the increased echoes during rainy weather are possibly false ones while during fine weather they are mostly from nearby echoes. Therefore, the count increases from criterion 1) to 2) and 1) to 3) are similar. In addition, the RB echoes increase more than the SB echoes on fine nights (except for the night of 20 June 2000 with criterion 4), while the SB echoes increase more on the rainy nights. These results suggest that interference from nearby echoes is not uncommon, even during low migration nights, and that the shorter the sample period for averaging signal intensity, the more sensitive the algorithm (SB signal uses 32 sample points for the average of signal strength).

3.1.2 Algorithm reliability Although the number of identified echoes increases as the criteria are made less demanding, a greater proportion of the echoes are of poor quality due to interference from nearby targets or of false echoes from rain droplets. Table 3-2 lists echo counts from RB samples for these five nights. The percentage of analysable echoes decreases as total echoes increase, indicating the unanalysable echoes increase by even greater amounts. This is evident for SB echoes [especially the change from echo-selection Table 3-2 RB Echo Counts with Different Criteria for Echo Delimitation 6dB Full 6dB Half 3dB Full 3dB Half Date Echo Quality Parabola Parabola Parabola Parabola G (%) 21.5 16.8 17. 9 15.0 A (%) 97. 8 86.3 95.9 83.6 00Jun23 U (%) 2.2 13.7 3.8 16.1 Total 223 315 291 366 G (%) 24.5 18.1 17.2 15.1 A (%) 97.6 89.0 94.5 87.2 00Jun20 U (%) 2.4 10.8 5.5 12.3 Total 2039 2963 2985 3508 G (%) 11.6 8.6 8.1 6.9 A (%) 97.0 91.3 94.5 90.7 00Jun21 U (%) 2.9 8.1 5.3 8.3 Total 8058 11953 12538 15515 G (%) 34.5 21.5 13.7 6.0 A (%) 98.8 84.0 91.4 82.9 00Jun19 U (%) 1.2 15.7 7.9 14.3 Total 168 312 430 1150 G (%) 18.1 12.2 9.4 4.2 A (%) 96.9 81.6 89.9 76.5 00May02 U (%) 2.1 14.3 7.8 21.1 Total 486 876 1027 2451 Data is same as in Figure 3.16. Total counts include the analysable, unanalysable (U) and short echoes.

110 Chapter 3: Implementation of IMR Signal Processing criterion 3) to 4)], as they are selected by ever more sensitive methods (i.e. shorter averaging period) (Figure 3.16). In fact, the number of good echoes varies little with the selection criteria on all five nights. This shows clearly that only a minority of echoes are clear and strong, and free of interference from echoes nearby. Therefore, making the criteria of echo delimitation less demanding hardly affects the number of good quality echoes, but allows more weak echoes to be selected.

Short echoes that are not long enough to analyse are the most affected by the change to the echo-delimitation criteria. Table 3-3 shows the numbers of these counted- only echoes from the different echo-selection schemes for the five nights. As expected, the less demanding the echo-selection criteria, the more short echoes are delimited. However, the number changes due to reducing the requirement for echo strength [from criterion 1) to 3)] are less than those from reducing the requirement for the completeness of the echo shape [from criterion 1) to 2)] in the case of RB samples, except for the observations under light rain. Under light rain, only strong echoes can be selected, so the interference from nearby echoes is very rare. For the SB samples, in contrast, the number increases are greater from the decrease of echo strength than from the reduced requirement for shape-completeness, except for the observations of heavy migration. This exception, again, illustrates that interference is common in the echoes from the high-migration night. Therefore, for echo delimitation, SB signals are sensitive to the strength requirement but RB signals to the shape completeness, due to its shorter averaging period for the signal strength.

Table 3-3 Counts of Short Echoes with Different Criteria for Echo Delimitation Peak Shape 6dB Full 6dB Half 3dB Full 3dB Half & Strength Parabola Parabola Parabola Parabola 00Jun23 RB echo 0 0 1 1 fine SB echo 0 0 0 1 00Jun20 RB echo 0 5 1 18 fine SB echo 0 3 4 35 00Jun21 RB echo 8 66 25 163 fine SB echo 5 56 44 176 00Jun19 RB echo 0 1 3 33 rainy SB echo 1 2 29 137 00Ma02 RB echo 5 36 24 59 rainy SB echo 0 0 15 73 Short echoes are 0.6 and 0.3 s long for RB and SB samples respectively. Data is same as in Figure 3.16. 3 Some Measures of IMR Performance 111

The echo-selection criterion 2), i.e. a 6 dB half parabola is required form an echo to be selected, has been adopted in the program DE, for the consideration of better handle of echo interference during heavy migration and of possible virga during light rain, as well as algorithm sensitivity to signal conditions and echo quality from analysable echoes. As the number is such a small proportion of the total targets, short counted-only echoes are not considered any more.

3.2 Measurement Accuracy

IMR hardware stability is essential for long-term operation (see Chapter 2). Hardware performance can to some extent be monitored from the observation results – any suspicious results may be an indication of a hardware fault. The reliability of the data processing is similarly critical for application of IMRs to long-term studies of insect migration.

3.2.1 Angular range of detection The IMR beam has an offset angle of about 0.2° to its zenith axis. The closest approaches of detected targets, therefore, cannot be too large, as targets traversing the radar beam far away from the zenith axis will produce very weak signal intensities from this pencil beam with ~1.1° beamwidth. Figure 3.17 shows the distribution of the closest approaches from all targets that were fully analysable under the RB observation mode. About 99.5% of 9385 targets were detected within an angular distance of 1.0 beamwidths (BW) from the zenith axis, with the maximum of 1.36 BW at the lowest range gate. Of these detected targets, half were detected within 0.2 beamwidths of the zenith. In addition, as the gate range goes higher, the measured closest approach gets narrower (Figure 3.17a). For the night with one dominant species, this is in a good agreement with the IMR beam shape.

3.2.2 Body alignment of target The target RCS can be measured under both RB and SB mode operations, though under the latter mode, only a lower limit can be estimated. With the RCS comparison between these two operation modes, the IMR beam alignment and all angular measures can be crosschecked. Figure 3.18 shows the distribution of the RCS coefficient a0 (d) and the

112 Chapter 3: Implementation of IMR Signal Processing

0.7 (a) 0.6 Median Mean 0.5

0.4

0.3

0.2 Beam Width

0.1

0.0

200 400 600 800 1000 1200 1400

6 (b) Range Height (m)

5 200-1300m, 19-02h, 11-12 Feb 1999, Bourke 4 (9385 targets)

Percentage (%) 1 > 1.10

0 0.0 0.2 0.4 0.6 0.8 1.0 Closest Approach to the Zenith (BW) Figure 3.17 Distribution of the Distance of Closest Approach to the Zenith The mean closest approach is plotted with one standard deviation against the height of range gate centre, along with the median value at each range gate (a). The histogram of the closest approaches (b) is from all analysable targets at all altitudes. Data is from the IMR observations at Bourke during the 8 hr following dusk on the night of 11-12 Feb 1999. orientation directions (f) obtained from the RB operation mode, in comparison with the distributions of the lower limit on the RCS (a-c) at three azimuthal directions obtained under the SB mode. The maximum RCS would usually be expected to occur when the radar beam polarisation is parallel to the insect’s longitudinal body axis (Drake et al. 2002b), and, a cross-check of this provides an indication that the analysis procedures are consistent. From the RB observations, the flying insect population had a collective orientation at 153 – 333° (Figure 3.18f) on this night. Thus the maximum measurement of the lower limit on the RCS should occur when the polarisation is aligned along this direction. For the three fixed-direction observations at 63.7, 123.7 and 183.7°, the latter two are equidistant from 153° while the first is at a right-angle to it (Figure 3.18a, b, c). The peaks of the lower limit on the RCS reflect this.

3.2.3 Target speed A cross-check of the speed estimates from the two operating modes also provides some validation of the parameter reliability. Target speed is estimated by different methods in 3 Some Measures of IMR Performance 113

500 o 0 O (e) 352 Displacement 400 (a) Polarisation 63.7-243.7 300 Total 4375 200 100 0

500 O 270 90 Polarisation 123.7-303.7 400 (b) 300 Total 4437 200 100 0

500 O 180 (c) Polarisation 183.7-3.7 400 333o 0 300 Total 4461 (f) Orientation 200

Number of Echoes 100 0 -4 -3 -2 -1 0 1 2 Lower Limit on RCS [log10(cm )] 270 90 6000 (d) Polarisation 0~360O 4000 Total 35425 >1.4 2000 0 153o -4 -3 -2 -1 0 1 180 RCS a (200-1400m, 19-05h, 28-29 Feb 2001, Bourke) 0 Figure 3.18 Histograms of RCS Measurements, Displacement and Orientation on the Night of 28 Feb 2000 at Bourke Histograms are for all analysable echoes obtained between 19 and 05 h, at 200-1400 m, for the lower limit on RCS at the three SB polarisation azimuths accordingly (a, b, c, 13273 SB echoes) and the polarisation average of RCS a0 under the RB-mode (d, 35425 RB echoes). Equal-area rose diagrams of displacements (e) and orientations (f) are plotted in 10° bins with mean directions (vectorial averages, see Appendix D.3) from the 35425 RB echoes.

14 200-1300m, 19-02h, 11-12 Feb 1999, Bourke 12 10 8 R Echo (11309) 6 S Echo ( 4959) 4

Percentage (%) Percentage 2 >40 0 0 5 10 15 20 25 30 35 40 Speed (m/s) Figure 3.19 Speed Comparison of RB and SB echoes Histograms of speeds are from all analysable targets detected by the IMR under both the RB and SB modes, at altitudes of 200-1300 m between 19-02 h on the night of 11-12 Feb 1999, Bourke. the RB and SB modes of IMR operation. Therefore, a speed comparison can be used as an indicator of measurement accuracy. Figure 3.19 shows such a comparison. The speed from the RB echoes is 10.36±0.04 (median 10.07) m/s, which is not significantly (two- sample t-test, P = 0.17) different from the speed from the SB echoes (10.40±0.05, median 10.26 m/s). The slight differences may arise from the value of the beam offset, which is very difficult to measure accurately.

114 Chapter 3: Implementation of IMR Signal Processing

Small insects can probably be used as tracers of the wind, as their air speeds are low (Schaefer 1976; Pedgley et al. 1982; Drake 1990a; Riley 1994; Riley & Edwards 1997). Therefore, a speed comparison with upper winds from a radiosonde ascent can validate IMR speed-estimation accuracy. Figure 3.20 shows a comparison of speeds of small and large insect targets from an obvious two-sized population. The small insects (10-40 mg) had speeds that were slightly faster than the upper winds, while the large ones (50-300 mg) had much higher speeds (~ 3 m/s faster). Two-sample t-tests show that the small-insect speeds are not significantly faster than the wind speeds (P = 0.08), while the large-insect speeds are significantly faster than both the winds (P = 0.02) and the small-insects (P = 0.0002). The differences between small-insect speeds and wind speeds observed here may be real, or perhaps a result of the different locations at which the two quantities were measured.

Temperature (°C) 0 5 10 15 20 25 1200

Air Temperature 1000 Upperair Wind Large Insects 800 Small Insects

600 Altitude (m)Altitude 400

200

0 02468101214 Speed (m/s) Figure 3.20 Speed Comparison between Small and Large Insects against Upper Winds Profiles of the mean speeds of small (10-40 mg) and large (50-300 mg) insects are from observations under the RB mode of the Bourke IMR during the period 21:00-00:35 h on the night of 16-17 Mar 1999. Wind speeds and temperatures from the radiosonde at 21 h at Cobar, which is about 160 km south of Bourke, are also included; the upper-air data are from the Australian Bureau of Meteorology.

3.2.4 Wingbeat frequency The ability to estimate wingbeat frequency from both the RB and the SB samples provides a further validation of the accuracy of the DE algorithm’s reliability. Figure 3.21 shows the distributions of wingbeat frequency from the SB and the RB echoes for a 3 Some Measures of IMR Performance 115 night when the detected insects were mainly of a single size class. Frequencies selected for comparison were between 14 and 40 Hz, as less than 3% of the RB echoes and 8% of the SB echoes gave higher frequencies on this occasion. The frequency distributions have a common peak at about 28-29 Hz and the same symmetric approximately normal distribution. About a quarter of the frequencies from RB echoes are not from real wingbeats, as there are strong narrow peaks at about 15 and 20 Hz that are most likely from harmonics of the 5-Hz conical scan. This interference from the conical scan does not interfere greatly with interpretation of the spectra, as most of the species detected by the IMR have wingbeat frequencies of over 20 Hz.

12 S Echo (4064) 10 R Echo (6929) 8

4 >40.5

Percentage (%) 2

0 15 20 25 30 35 40 Wingbeat Frequency (Hz) Figure 3.21 Comparison of Distributions of Wingbeat Frequencies from SB and RB echoes Histograms of the strongest frequencies between 14 and 40 Hz for SB and RB echoes observed at Bourke at altitudes of 200-1300 m between 19-05 h on the night of 11-12 Feb 1999.

4 Conclusions

The new echo-delimitation algorithm has overcome the deficiency of the original procedure on delimiting echoes from the samples of heavy insect-flight activities or rainy weather. Although the algorithm does not work well when samples are taken at rainy weather, especially for SB echoes with the least demanding echo- selection criterion, the rain detector incorporated with the IMR will detect most rains and the DE will be instructed to skip these samples. The only exception is if rain drops do not reach the ground, i.e. virga, the rain detector will not detect them. The current algorithm has no means of checking if an echo is from a rain drop. Recognising that there must be a trade-off between the algorithm sensitivity and overall quality of delimited echoes, the criterion 2) has been chosen for DE routine processing.

116 Chapter 3: Implementation of IMR Signal Processing

The program DE has been developed to automatically process IMR observations for each whole night. It delimits target echoes from background noise and adjoining echoes, and extracts the trajectory and identification parameters of the targets, all without user interaction. With the new echo-delimitation algorithm, it remains effective even if the level of background noise varies markedly, and when the migration intensity is high. The one-way echo-search scheme has improved the computation efficiency. Incorporation of weighted least-squares fitting with the new Golub-Reinsch algorithm of singular value decomposition, and of spectral analysis with the windowed radix-2 fast Fourier transform, have improved the speed and stability of signal processing, as the use of extended memory. On the 486 DX2/66 PC, DE is also capable of processing data efficiently and quickly so that results can be available within a few hours of a migration occurring, normally late the next morning.

Estimating wingbeat frequency from rotary-beam echoes creates a direct link between wingbeat values and estimates of the body size and body shape. Hence it should be very useful in the identification of the targets detected during radar observations. The success of retrieving wingbeat frequency from RB echoes has also demonstrated that with DE the high accuracy can be achieved on parameter estimation.

In summary, the DE echo analysis procedures, which have been used to produce all the datasets in the remainder of the thesis, produce good results and provide a measure of echo quality. It has proved very reliable and stable, and not subject to unexpected termination.

4 Characterisation of IMR Targets

1 SIZE OF AN INSECT TARGET ...... 118 1.1 MASS ESTIMATORS...... 119 1.2 RELIABILITY OF RCS MEASUREMENT BY THE IMR...... 121 1.2.1 Measurement accuracy for target size ...... 122 1.2.2 Comparison of mass estimates ...... 123 2 SHAPE OF AN INSECT TARGET...... 126

2.1 RATIO OF aa42...... 126

2.2 RATIOS OF aa20 AND aa40...... 128

2.3 RATIO OF σ yyσ xx ...... 129 2.4 EXAMPLES OF RCS SHAPE FACTORS ...... 132 3 WINGBEAT FREQUENCY OF AN INSECT TARGET...... 135 3.1 TEMPERATURE DEPENDENCE OF WINGBEAT FREQUENCY ...... 136 3.2 WINGBEAT MODULATION PATTERN ...... 138 3.3 WINGBEAT FREQUENCY AND INSECT SIZE AND SHAPE...... 139 4 IDENTIFICATION OF CHORTOICETES TERMINIFERA FROM IMR ECHOES ...... 141 5 DISCUSSION...... 143

Target identification has long been an unsolved problem in radar entomology. Without direct aerial sampling or ancillary information on the migrating populations, the species cannot be determined and thus the IMR observations cannot be applied to the practice of insect pest management. The new signal-processing program (described in Chapter 3) allows characteristic information on the size, shape, wingbeat frequency and modulation pattern of the flying targets to be retrieved from the signals digitised under the two modes of IMR operation. It would be ideal to measure the radar signatures of dominant migrant species from the study region at laboratory using the equipment with the same IMR configuration and use these values to match the IMR observations for target identification. The lack of this laboratory facility and the remote locations of IMR sites made this study infeasible. To explore the possibility of directly identifying targets from the signal, echoes were examined from a number of heavy migration events detected by

117 118 Chapter 4: Characterisation of IMR Targets the IMR at Bourke. By drawing on light-trapping and field-survey results conducted by the Australian Plague Locust Commission (APLC), the most likely migration nights of Australian plague locusts, Chortoicetes terminifera, were identified and their echo characteristics were determined. These criteria were then used to identify the migration events of C. terminifera, which formed the basis of the studies of this species’ migration and orientation behaviours in the final chapters of this thesis, on the both IMR sites.

1 Size of an Insect Target

Target size, which can be inferred from the back-scattering radar cross-section (RCS) (Blake 1986), is a direct indicator of the category of target species. In the Rayleigh scattering region, the RCS increases at the sixth power of target diameter, and a big RCS therefore indicates that the target mass is large. Most insects, typically less than 200 mg (Riley 1985; Aldhous 1989), fall into this region at X-band frequencies. Since it is the water in the insect body that reflects the radar waves, an initial approach to the estimation of insect RCS was to determine the size of the water drop with the same water content as the insect, and then use the calculated RCS for this insect (Rainey 1955). However, it has since been found that the insect’s RCS is underestimated and the relationship cannot reliably be used to estimate target mass (Riley 1973; Aldhous 1989). In the Mie (resonant) region which very large insects (>1000 mg) occupy at X-band, the RCS does not always increase with the mass of the target; instead, it varies in a complex way with target size, aspect angle and shape. Therefore, a small RCS may not mean the target size is small. Riley (1985) found that the RCS of big (2100 – 3500 mg) Desert Locusts is at its maximum when their body axis is perpendicular (rather than parallel) to the electric vector at X-band wavelength, and later Aldhous (1989) confirmed this finding from a larger sample. Fortunately, there are not many known migratory insects of this size in Australia, so the possibility of targets being in the Mie region can be mostly ignored in this study. Birds will be in either the Mie or geometric. Migration intensities of birds in inland Australia are not as intense as in the northern hemisphere continents and birds’ echoes have not often been detected by Australian entomological radars (VA Drake, 2001, personal communication). Therefore, birds have been excluded as a significant source of echoes throughout this study. 1 Size of an Insect Target 119

1.1 Mass Estimators

Several different expressions have been developed for estimating insect mass from RCS measurements, but none is entirely satisfactory. Previous estimations are reviewed here and a new equation, used in this study, is then introduced.

Aldhous (1989) has made extensive RCS measurements in the laboratory and concluded that insect mass m (in mg, over the range 40 – 4000 mg) can be estimated 2 from the polarisation-averaged RCS a0 (in cm ) as Equation Chapter 4 Section 1

logmaa=− 4.24 4.17 − 3.82log00 (0.1 << 10) 4.1 which is plotted as Figure 4.1a, or alternatively as

2 logm =+ 2.54 0.766logσσσyy + 0.179(log yy ) (0.01 << xx 10) 4.2

2 where σ xx and σ yy (both in cm ) are defined to be, respectively, the maximal RCS of the insect when the polarisation is rotated and the RCS when the polarisation is perpendicular to this direction (Aldhous 1989). σ xx and σ yy are calculated as

⎪⎧σ xx = aaa024++ ⎨ ⎩⎪σ yy = aaa024−+

where a0 , a2 and a4 are RCS coefficients (Smith & Riley 1996).

Riley (1992) published another empirical mass estimator (Figure 4.1b)

105 m =<σσ( 0.1) 4.3 6.4 xx xx

2 For big insects with σ xx ≥ 0.1 cm in the Mie region, Equation 4.2 was adopted by Smith et al. (1993). However, these two estimators leave a gap between 50 and 80 mg. A more recent re-examination (Chapman et al. 2002a) has found that equation 4.3 gives 2 accurate estimates only when σ xx ≤ 0.0032 cm , and that for larger σ xx mass is best

2 estimated with a0 instead: when a0 < 0.25 cm ,

105 ma= 4.4 6.4 0

120 Chapter 4: Characterisation of IMR Targets

is used; and when a0 ≥ 0.25 ,

2 logmaa=+ 2.205 0.8729log00 + 0.3323(log ) 4.5 is adopted (Figure 4.1b). Equation 4.3 is plotted in Figure 4.1 on the assumption that a0 = 0.562σ xx , i.e. σ xx≈ 5σ yy (Riley & Smith 1999; Riley et al. 2003). It is claimed that these equations provide a mass estimate with an error of less than a factor of two in a mass range of three orders of magnitude (Chapman et al. 2002a). However, this three- region relation may produce a false mass peak between 5 and 7 mg, as the two mass 2 estimators for small insects may be discontinuous when σ xx is around 0.0032 cm since the ratio of σ xx to a0 varies from insect to insect.

10000 4.24-(4.27-3.82 loga )0.5 (a)Y =10 0 (0.1

_ 2.5 0.325 (c)Y =σ /e (σ >0.262) h x _ 1/3.12 10.12/3.12 Y =σ /e (σ ≤0.262) h x a (Aldhous,1989) 10 0 Mass (mg) (Russell & Wilson, 1997) a (Riley et al,2003) 0 a (Wolf et al, 1993) _0 σ (Russell & Wilson,1997) h 1 (d) 2.2266+0.89424 loga +0.21932 (loga )2+0.03039 (loga )3 Y = 10 0 0 0 R2 =0.92272, SD =0.1877, P<0.0001 (Fitted from Aldhous,1989 & Riley,1992) 0.1 1E-6 1E-5 1E-4 1E-3 0.01 0.1 1 10 100 Polarisation-Averaged Back-Scatter Radar Cross-Section a (cm2) 0

Figure 4.1 Variation of Mass with the Polarisation-Averaged Radar Cross-Section a0 Points are plotted from laboratory measurements at X-band wavelength (3.2 cm) (Aldhous 1989; Wolf et al. 1993a; Riley et al. 2003), or scaled to X-band (Russell & Wilson 1997).

Russell and Wilson (1997) collected all published RCS measurements of insects and birds, scaled them to X-band frequency if they were measured at other frequencies, and calculated aspect averages (for horizontal polarisation) σ h . While aspect averages have no simple relation to polarisation averages, it seems likely they will be of similar magnitudes. Some masses were also estimated from body lengths and widths when the 1 Size of an Insect Target 121 weights were not provided in their sources. The relationship of mass (g) and RCS (cm2) was found to be

10.12 3.12 ⎪⎧σ h =≤em( L 10mm) 4.6 ⎨ 0.13 0.40 ⎩⎪σ h =>em( L 10mm)

where L is the body length of the insect and σ h is the aspect-averaged RCS at X-band, with a goodness-of-fit R2 = 0.91 (Figure 4.1c).

The mass estimators from Smith et al. (1993) have not been adopted for use with the insects detected by the IMRs, because of the gap in the mass distribution produced by the discontinuity between Equations 4.3 and 4.2. However, the polarisation-averaged

RCS a0 appears a good candidate for estimating mass over the wide range of RCS values in IMR detection. Accordingly, a 3rd-order polynomial (Figure 4.1d) for mass as a function of a0 (both logarithmically transformed) has been fitted to the measurements by Aldhous (1989) and two additional data points for brown planthopper Nilaparvata lugens measured by Riley (1992), these data being selected because they appear the most reliable and cover a wide mass range. The result is

23 logmaaa=+ 2.2266 0.89424log00 + 0.21932(log ) + 0.03039(log 0 ) 4.7

( n = 53, R2 = 0.92 and SD = 0.19 at P < 0.0001). The reason for using this equation is not based on theory but rather on the need for a simple calculation and a smooth mass distribution. It overcomes the overestimates that arise with the parabolic fits for very weak echoes, and the difficulties caused by transitions between σ xx , σ yy and a0 . This mass estimator produces reasonably good resolution for separating insect targets of the wide range of masses detected by the IMRs in inland Australia. For the IMR at Bourke, the smallest targets detectable, in the range 200 – 250 m above ground level, had estimated mass of 0.6 mg.

1.2 Reliability of RCS Measurement by the IMR

Because of the difficulty of positioning a known target high over an IMR, it is impractical to perform a direct calibration, nor aerial sampling on the remote IMR sites. Thus, the estimates of target sizes were crosschecked with ancillary information.

122 Chapter 4: Characterisation of IMR Targets

1.2.1 Measurement accuracy for target size Although the RCS of a target is related to the radar wavelength and other properties, i.e.

34 4 prr(4π ) dpd σ ==22 4.8 ptrgCλ

(see Equation 3.1), for a specific radar, the term C in Equation 4.8 is a constant. The received power pr from a target falls off as the fourth power of the range d. This relationship can be used to validate both the radar performance and the algorithm for extracting RCS values, if observation of known targets can be made (Riley 1978). However, because of the target positioning difficulty, it has been necessary to rely only on naturally occurring targets.

-40 Average -45 Median -50

-55

-60

-65 Y=10 [4.93604 - 3.94487 log(X)] -70 R = -0.99683, SD = 0.08739, P < 0.0001

Received Power (dBm) Power Received -75

-80 200 400 600 800 1000 1200 1400 Altitude (m) Figure 4.2 Sensitivity of the IMR at Bourke Mean and median values of received mean power are for 10550 targets (5% extreme values are removed from each gate range) detected at 8 altitudes between 20 h and 01 h on the night of 11-12 Feb 1999.

A night of heavy migration, 11 – 12 February 1999, when the airborne insect population was predominantly a single species, has been selected to validate the IMR’s sensitivity to target sizes and ranges. The dominance of a single species was indicated by the highly concentrated distribution of wingbeat frequencies (see Figure 4.15a) of the detected insect targets at all altitude levels on this night and by APLC survey records that indicated that adult Australian plague locusts were abundant in the wide region. Figure 4.2 shows the relationship of the received power with the target range. The logarithmic-transformed power average from the detected targets at each gate is fitted with a linear regression against the logarithmic-transformed gate range. The fitted line has a slope of −±3.94 0.129 (n = 8 , R = −0.997 , P < 0.0001), which is very close to the expected inverse-fourth power of range dependence. The median values of received power are very close to those average values, confirming similar sizes of target were 1 Size of an Insect Target 123 detected by the IMR. This provides a good indication that both the radar hardware and the analysis software were functioning properly.

1.2.2 Comparison of mass estimates Masses for newly fledged adults of Chortoicetes terminifera were obtained from two populations laboratory-reared at the APLC in March and April 2001 respectively. Specimens were killed and weighed immediately (DM Hunter, 2001, personal communication). Comparisons between sexes and samples were made with the two- sample independent t-test (Sincich 1993) (Table 4-1). At the significance level α = 0.1, there are no significant differences between the two samples for males and for females, even though their ages were slightly different, but differences exist between the sexes in both samples. Thus, the two samples are pooled by sex (Table 4-2), and the average mass of these young adults would be 0.213 g (median 0.203 g), given the sex ratio 1:1. As the sex ratio of a field population may vary at different times or locations (Clark et al. 1969), the mass average of a population depends on its population structure.

Table 4-1 Two-Sample Independent t-Test of Masses of Young Chortoicetes terminifera Adults Sample N Mean (g) SD SEM t-Value DoF P-Value

First 12 0.162 0.070 0.020 Male 0.306 25 0.762 Second 15 0.156 0.014 0.004 First 12 0.248 0.058 0.017 Female -1.682 28 0.104 Second 18 0.280 0.044 0.010 Male 12 0.162 0.070 0.020 First -3.294 22 0.003 Female 12 0.248 0.058 0.017 Male 15 0.156 0.014 0.004 Second -10.387 31 <0.0001 Female 18 0.280 0.044 0.010 Data is from D M Hunter (APLC); samples taken in March and April 2001. SD—Standard Deviation, SEM—Standard Error of Mean, DoF—Degree of Freedom.

Table 4-2 Masses of Young Chortoicetes terminifera Adults Sex Mean (g) SD SEM Median Male 0.159 0.04698 0.00904 0.150 Female 0.267 0.05162 0.00943 0.255 Average 0.213 0.203 Data as in Table 4-2, the average of both sexes is based on the assumption that sex ratio is 1:1.

124 Chapter 4: Characterisation of IMR Targets

The masses of fresh specimens of C. terminifera caught in light traps during a migration period were in the region 133 – 350 mg for 363 males and 324 – 860 mg for 181 females (Reid et al. 1979). Since these locusts were caught during migration, they must have accumulated some of the fat reserves required for migration (Hunter 1981a), and some of the females may have started egg development. Therefore, it is not surprising that they were heavier than these newly fledged adults.

20 10 1250-1300m, 316 0 30 20 1100-1150m, 522 10 0 40 20 950-1000m, 728 0 60 40 800-850m, 1016 20 0 80 650-700m, 1546 Target Number 40 0

40 500-550m, 1765 0 80 40 350-400m, 2249 0 160 >500 80 200-250m, 2650 0 0 100 200 300 400 500 Mass (mg) Figure 4.3 Mass Distribution of Targets on the Night of 11-12 Feb 1999 at Bourke 10792 analysable targets detected by the IMR between 20 h and 01 h.

Direct RCS measurements of C. terminifera are available for only two adult individuals, a male of 250 mg and a female of 470 mg with aspect-averaged RCS of 0.6 cm2 and 0.4 cm2 respectively at X-band radar (Schaefer 1976). Their mass estimates would be 110 and 80 mg for the male and female respectively, if the measured RCSs were treated as polarisation-averaged a0 and substituted into Equation 4.7. The locust masses are obviously underestimated by this equation in this instance. The target population detected at Bourke on the night of 11 – 12 Feb 1999 has masses estimated predominantly in the range 50 – 250 mg ( > 70% targets), with the median at 130 mg and long right tails (Figure 4.3). The masses from the insects detected at the altitude 200 – 250 m were significantly different from those at higher levels, so perhaps these were 1 Size of an Insect Target 125 some smaller species. If the sizes of detected locusts on this night were young adults with little fat reserves, their mass estimates would be right. Otherwise, their masses would be underestimated, possibly by half, if they were similar to the sizes of these two measured specimens. However, due to the lack of aerial sampling facility, it is unknown whether the airborne population on this occasion was very young. Another possibility is that mass estimate is particularly difficult at mass typical of locusts because the size of these insects puts them in the resonant region at X-band wavelength and here RCS varies rather erratically with target size (Riley 1985).

Figure 4.4 shows distributions of estimated masses for samples of insect targets from different seasons. Single species were predominant on the nights of 11 – 12 February 1999 and 16 – 17 August 1998 (Figure 4.4a, d), but a more varied fauna was present on the nights of 05 – 06 November 1999 and 15 – 16 September 1998 (Figure 4.4b, c). The peak masses on the nights of 11 – 12 February 1999 and 05 – 06 November 1999 are similar, indicating similar-sized insects were predominant on these two nights.

(a) 200-1300m, 20-01h, 10 (b) 350-1400m, 00-04h, 8 11-12/02/1999, Bourke 06/11/1999, Bourke (10792 targets) 8 (2844 targets) 6 6 4 4 >800 >640 2 2

0 0 0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 18 (c) 18 (d) 16 (%) Probability 200-550m, 19-22h, 200-400m, 19-22h, 14 15/09/1998, Bourke 16 16/08/1998, Bourke 14 12 (5023 targets) (3393 targets) 12 10 10 8 8 6 6 >30 4 4 >20 2 2 0 0 0 5 10 15 20 25 30 0 5 10 15 20 Mass (mg) Figure 4.4 Mass Distributions of Insect Targets from Different Seasons Counts are from all analysable targets.

126 Chapter 4: Characterisation of IMR Targets

2 Shape of an Insect Target

It has been suggested that angular variations of the RCS resulting from the effect of body shape may be useful in the identification of insect targets (Riley 1978). The variation of RCS with polarisation (Riley 1973), provides one such source of information about body shape. From the RCS parameters, a0 , a2 , and a4 , that are routinely estimated from the IMR rotary-beam signals (see Chapter 3), a number of alternative means of target shape can be derived (Dean & Drake 2005).

2.1 Ratio of aa42

For a target at fixed aspect and with linearly polarised radiation, the RCS σ varies with polarisation angle φ (see Chapter 3) as

σ ()φφθφθφπ=+aa02 cos2( − 1 ) + a 4 cos4( − 2 ) = 0~

where θ1 defines an alignment angle for the pattern and θ21−θ provides an indication of the degree of body asymmetry (Aldhous 1989; Hobbs & Aldhous 2006). When the ratio of RCS coefficients

a 1 4 ≤ , 4.9 a2 4

σ xx , the maximum value of RCS, occurs when the polarisation is parallel to the longitudinal body axis of the insect, and σ yy , the minimum, occurs when the polarisation is perpendicular to this axis (Aldhous 1989). This indicates that the insect length is much smaller than the radar wavelength (Schaefer 1976), and the target is thus in the Rayleigh region. When

a 1 4 > 4.10 a2 4

σ xx and σ yy are primary and secondary maxima, and the RCS minima are thus located between σ xx and σ yy angles. As the insect size increases, σ xx eventually occurs when 2 Shape of an Insect Target 127 the polarisation is perpendicular to the longitudinal body axis of the insect (Aldhous 1989); this occurs when the insect’s dimensions are comparable to the radar wavelength

(Riley 1985). Figure 4.5 shows some examples of aa42 distributions from the IMR observations at Bourke. The median values of aa42 were about 0.35 for 11-12 February 1999, 0.325 for 06 November 1999, 0.2 for 15 September 1998, and 0.225 for 16 August 1998. The insects detected on the former two nights are therefore long and thin, but might be slightly different in size or shape; the latter are relatively small but wide. Although there is no simple relationship between aa42 and insect size or axial body ratio (Aldhous 1989), nonetheless, the specific ratio of aa42 may indicate the shape character of an insect species.

10 15 (a) 200-1300m, (b) 350-1400m, 8 20-01h, 00-04h, 11-12/02/99, 06/11/99, Bourke 10 Bourke 6 (10217 targets) (2291 targets)

4 5 >1.6 2 >1.6

0 0

10 (c) 200-550m, 15 (d) 200-400m,

Probability (%) Probability 19-22h, 19-22h, 8 15/09/98, 16/08/98, Bourke 10 Bourke 6 (4751 targets) (3098 targets)

4 5 >1.6 2 >1.6

0 0 0.0 0.4 0.8 1.2 1.6 0.0 0.4 0.8 1.2 1.6 Ratio of a /a 4 2

Figure 4.5 Distributions of aa42 for Insect Targets from Different Seasons Data are from all analysable echoes, from single nights in a) summer (mass range of 17-825 mg), b), c) spring (50-650 mg and 2-30 mg respectively), and d) winter (2-20 mg). The nights are the same as those in Figure 4.4.

128 Chapter 4: Characterisation of IMR Targets

2.2 Ratios of aa20 and aa40

The RCS coefficients a0 and a4 generally increase, while a2 varies in a complicated way, as the insect mass increases. No simple relationships have been found among them

(Aldhous 1989). However, the values of a2 and a4 in relation to a0 may be species dependent. Figure 4.6 shows distribution patterns of the ratio of aa40 against aa20 for insects detected in different seasons. The major insect species detected on the nights of 11 – 12 February 1999 and 05 – 06 November 1999 had similar sizes and both had long and thin bodies, but their shapes appeared to be slightly different from the ranges and concentrations of aa20 distributions. The small insects (Figure 4.6c, d) had obviously different ranges of their parameters. The value of a2 is more diverse in some insects (Figure 4.6a, c) than others, which may be due to variations within the insect population or to varying sensitivity of the polarisation pattern to insect shape.

200-1300m, 20-01h, 11-12/02/1999, Bourke 350-1400m, 00-04h, 06/11/1999, Bourke 0.6 300 120 246 98.4 (a) 192 (b) 76.8 0.5 138 55.2 84.0 33.6 0.4 30.0 12.0

0.3

0.2 0 /a

4 0.1

200-550m, 19-22h, 15/09/1998, Bourke 200-400m, 19-22h, 16/08/1998, Bourke 0.6 100 100

Ratio of a Ratio 82.0 82.0 (c) 64.0 (d) 64.0 0.5 46.0 46.0 28.0 0.4 10.0 28.0 10.0 0.3

0.2

0.1

0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 Ratio of a /a 2 0

Figure 4.6 Distributions of aa40 with aa20 for Insect Targets from Different Seasons Data sets and mass ranges are as in Figure 4.5. Colour scale shows the distribution of target counts from the spatial Kriging extrapolation. 2 Shape of an Insect Target 129

2.3 Ratio of σ yyσ xx

Aldhous (1989) found that the ratio of σ yy to σ xx has no simple relationship to the ratio of body length to abdomen width, a result that contradicts the earlier findings of

Hajovsky (1966). Figure 4.7 shows the RCS ratio of σ xxσ yy against the ratio of insect body length and diameter. A high correlation of R2 = 0.97 at P < 0.0001 was found in the data of Hajovsky (1966) after exclusion of the obvious outlier Tipula simplex which has a body length to diameter ratio of 13:1. However, this relationship does not exist in the study of Aldhous (1989), who measured 54 specimens from 18 species.

100 Hajovsky (1966) Aldhous (1989) 80 Y = -0.44567+0.98576 X+0.27147 X2

yy 2

σ R =0.97053, SD =1.97885, P<0.0001 60 to xx σ 40

Ratio of Ratio 20

Timpula simplex 0

0 2 4 6 8 10 12 14 Ratio of Body Length to Body Diameter

Figure 4.7 Relationship of σ xxσ yy to Length Diameter of Insect Targets A parabola is fitted to the data from Hajovsky (1966) (the range crane fly Tipula simplex is excluded from the fit). The plots also include the data from Aldhous (1989).

Since the RCS coefficient a2 changes complexly in relation to insect size and shape, the geometric ratio of an insect cannot be simply inferred from the RCS ratio

σ xxσ yy (Aldhous 1989). Figure 4.8 shows some distributions of the RCS ratio

σ yyσ xx from insect targets detected by the IMR at Bourke in different seasons. The somewhat similar distribution patterns from the large insects detected on the nights of 11 – 12 February 1999 and of 05 – 06 November 1999 may indicate these insects have similar body characters (Figure 4.8a, b), and these distributions are clearly very different from small insects (Figure 4.8c, d).

130 Chapter 4: Characterisation of IMR Targets

(a) 200-1300m, 18 (b) 350-1400m, 10 20-01h, 16 00-04h, 11-12/02/99, 14 06/11/99, 8 Bourke 12 Bourke (10217 targets) (2291 targets) 6 10 8 4 6 4 2 2 0 0 10 (c) 200-550m, 35 (d) 200-400m, 19-22h, 19-22h, 8 30 Probability (%) Probability 15/09/98, 16/08/98, Bourke 25 Bourke 6 (4751 targets) (3098 targets) 20

4 15 10 2 5 0 0 0.00.20.40.60.81.00.00.20.40.60.81.0 Ratio of σ /σ yy xx

Figure 4.8 Distributions of σ yyσ xx for Insect Targets from Different Seasons See Figure 4.5 for the details of the data sets.

When the insect mass increases, σ yy increases, while σ xx increases in small insects, but varies non-monotonically in large insects when the body length is comparable to the radar wavelength (i.e. in the Mie region). Figure 4.9a shows the relationship between this RCS ratio and mass from Aldhous’ data. No simple relation has been found, though σ opσ , which is the RCS ratio when the polarisation is orthogonal to and parallel to the longitudinal axis of the insect’s body respectively, seems to increase as the insect mass increases for insects larger than 300 mg. This may be due to limitations of the sample size, as there were no specimens in the mass ranges <45 mg and 650 – 1000 mg, and the very large insects ( > 1000 mg) were all Desert Locusts, Schistocerca gregaria Forsk. In addition, the IMR cannot tell if the polarisation is parallel or perpendicular to the longitudinal body axis of the insect target, so use of σ opσ is not practicable.

Figure 4.10 presents some examples of the variation of RCS ratio σ yyσ xx with mass for samples of echoes from the Bourke IMR. As with the previous parameters considered (Figure 4.5, Figure 4.6 and Figure 4.8), the patterns from the nights of 11–12 February 1999 and of 05 – 06 November 1999 are similar, but slightly different. 2 Shape of an Insect Target 131

1 (a) xx σ / yy σ 0.1 Ratio of Ratio 0.01 10 (b) p σ /

Figure 4.9 Differential RCS in Relation to Insect Mass

The ratio σ opσ (b) is equal to σ yyσ xx (a) in small to medium insects ( < 648 mg), but equal to

σ xxσ yy in large insects ( ≥ 648 mg). Data is from Aldhous (1989).

200-1300m, 20-01h, 11-12/02/1999, Bourke 350-1400m, 00-04h, 06/11/1999, Bourke (a) 70.0 (b) 35.0 0.8 57.4 28.7 44.8 22.4 32.2 16.1 0.6 19.6 9.80 7.00 3.50

0.4

0.2 xx σ / yy σ 100 200 300 400 500 600 700 800 100 200 300 400 500 600 200-550m, 19-22h, 15/09/1998, Bourke 120 200-400m, 19-22h, 16/08/1998, Bourke 260

Ratio of Ratio (c) (d) 0.8 98.4 213 76.8 166 55.2 120 0.6 33.6 72.8 12.0 26.0

0.4

0.2

5 10152025 4 6 8 10 12 14 16 18 Mass (mg)

Figure 4.10 Distributions of σ yyσ xx against Mass for Insect Targets from Different Seasons Frequencies are shown as equal-number contours. See Figure 4.5 for details of the data sets.

132 Chapter 4: Characterisation of IMR Targets

2.4 Examples of RCS Shape Factors

With shape measurements, targets of the same species are expected to have similar shape characters. For the IMR at a specific location, locally abundant insects and migrants could be detected repeatedly in the same season, and therefore, after some initial field surveys or light trapping, historical matching could be used to identify the insects from their RCS characters. Figure 4.11 and Figure 4.12 present three nights when the RCS measurements suggest that the same insect species was predominant.

From comparisons of masses and of the shape factors aa42 and σ yyσ xx , and the correlations amongst these, it was evident that the populations of migrating insects from the nights of 26 – 27 February 1999, 26 – 27 February 2000 and 03 – 04 March 2001 were very similar. The main difference is in the mass distributions, which may perhaps be due to difference in age structure, or conditions under which the populations had developed.

14 (a) (b) (c) 350-1000m, 21-02h, 12 26-27/02/99, Bourke 10 10 10 (6181 targets) 8 6 5 5 4 >1.0 >600 2 0 0 0 15 10 15 (d) (e) (f) 650-1400m, 21-04h, 26-27/02/00, Bourke 10 10 (5078 targets) 5

5 5 >600 >1.0

Probability (%) Probability 0 0 0 15 (g) 15 (h) 10 (i) 500-1400m, 19-02h, 04-05/03/01, Bourke (9119 targets) 10 10 5 5 5 >1.0 >600

0 0 0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 100 200 300 400 500 600 Ratio of a /a Ratio of σ /σ 4 2 yy xx Mass (mg)

Figure 4.11 Distributions of Shape Factors aa42, σ yyσ xx and of Mass for Three Nights of Heavy Migration during Summer

Histograms of the ratios aa42 (a, d, g), σ yyσ xx (b, e, h), and of mass (c, f, i) are for the nights of 26- 27/02/99 (a, b, c, 7726 analysable targets), 26-27/02/00 (d, e, f, 5670 targets), and 04-5/03/01 (g, h, i, 10186 targets), all at Bourke. Small ( < 60 mg) and large ( > 800 mg) insects have been excluded. 2 Shape of an Insect Target 133

(a) 350-1000m, 260 (b) 650-1400m, 280 (c) 500-1400m, 500 0.8 21-02h, 213 21-04h, 230 19-02h, 410 26-27/02/99, 166 26-27/02/00, 179 04-05/03/01, 320 0 Bourke 120 Bourke 129 Bourke 230 /a 0.6 72.8 78.4 140 4 26.0 28.0 50.0

0.4

Ratio of a of Ratio 0.2

0.40.81.21.6 0.4 0.8 1.2 1.6 0.4 0.8 1.2 1.6 Ratio of a /a 2 0

(d) 60.0 (e) 50.0 (f) 90.0 0.8 49.2 41.0 73.8 38.4 32.0 57.6

xx 27.6 23.0 41.4 σ

/ 0.6 16.8 14.0 25.2

yy 6.00 5.00 9.00 σ 0.4

Ratio of of Ratio 0.2

200 400 600 200 400 600 800 200 400 600 Mass (mg)

Figure 4.12 Pattern Comparison of Shape Factors aa40 to aa20 and of σ yyσ xx to Mass

Equal-frequencies are plotted as contours of aa40 to aa20 (a, b, c) and σ yyσ xx to mass (d, e, f). Data sets are the same as in Figure 4.11.

The shape characters of RCS should be used in combination when identifying IMR targets, to maximise the chance of discriminating within the diverse fauna of migrating insects and because of the lack of any clear relationships between shape factors and mass. Figure 4.13 and Figure 4.14 show an example of a mix-sized population that was detected by the Bourke IMR in spring. The distribution of mass indicates two-sized groupings (Figure 4.13a), but both the small and the large insects had a similar distribution of aa42, with the peak values at 0.16 and 0.14 respectively

(Figure 4.13b). The ratios of σ yyσ xx show similar distributions, but for the small insects were highly concentrated (Figure 4.14e, f). Both of these patterns are not as clear as in the previous examples. However, their aa20 values had different distributions, with peak values about 0.9 for small insects and 0.5 for large ones (Figure 4.14b, c).

Thus, in this example, aa20/ appears to have discriminating power while σ yy /σ xx does not. Because these shape factors are interrelated, aa20/ and aa40/ contain more information about the target size and shape than aa42/ and thus should be used primarily.

134 Chapter 4: Characterisation of IMR Targets

12 (a) 16 (b) 200-1100m, 19-02h 14 19-20 Oct 2000, Bourke 10 12 10 8 Small Insects (2-40mg, 2008) 8 Large Insects (70-450mg, 3423) 6 4 >1.2

6 (%) Percentage 2 0 0.0 0.2 0.4 0.6 0.8 1.0 1.2 4 Percentage (%) Percentage Ratio of a /a 4 2

2 >450

0 0 100 200 300 400 Mass (mg)

Figure 4.13 Distributions of Mass and of aa42 for a Mixed Population Distribution of mass (a) has a peak at 18 mg and a broad rise around 130 mg (6385 targets). Distribution of aa42 (b) has peaks at 0.16 for small insects (2 – 40 mg) and at 0.14 for large ones (70 – 450 mg).

(a) 260 (b) 80.0 (c) 180 0.8 213 65.6 148 166 51.2 115

0 120 36.8 82.8

/a 0.6 72.8 22.4 50.4 4 Large 26.0 8.00 18.0 Insects 0.4

Ratio of a Small 0.2 Insects

0.4 0.8 1.2 1.6 0.4 0.8 1.2 1.6 0.4 0.8 1.2 1.6 Ratio of a /a 2 0

(d) 110 (e) 35.0 (f) 50.0 0.8 90.2 28.7 41.0 70.4 22.4 32.0

xx 50.6 16.1 23.0 σ

/ 0.6 30.8 9.80 14.0

yy 11.0 3.50 5.00 σ 0.4

Ratio ofRatio 0.2

100 200 300 400 5 101520253035 100 200 300 400 Mass (mg) Figure 4.14 Shape Characters of a Mixed Population

Distributions of (a) aa40 with aa20 and of (d) σ yyσ xx with mass are plotted as equal-frequency contours for the targets detected at 200 – 1100 m between 19 – 02 h (same as in Figure 4.13). (b) and (e) are for small insects of 2 – 40 mg. (c) and (f) are for large insects of 70 – 450 mg. 2 Shape of an Insect Target 135

3 Wingbeat Frequency of an Insect Target

Wingbeat frequency has been used by radar entomologists as the primary indicator of the identify of radar targets in the past (Schaefer 1976; Riley & Reynolds 1979; Drake & Farrow 1983). Wingbeating seems to always occur in nocturnally flying insects (Schaefer 1979), which have rarely been detected gliding as seen in some birds and in Desert Locusts in daytime swarms (Schaefer 1967; Roffey 1969; Rainey 1989). One possible instance was reported by Riley (1979), when targets thought for other reasons to be insects showed no wingbeat modulation. However, overlapping of wingbeat- frequency ranges between species, and the wide range of frequencies among individuals of the same species, reduces the value of this character for target identification (Riley 1978). Temperature dependence (Schaefer 1969; Drake et al. 2002b) extends the overlap. Nevertheless, it still has value for identification, especially when where are only a few candidate species.

136 Chapter 4: Characterisation of IMR Targets

3.1 Temperature Dependence of Wingbeat Frequency

Individuals of the same species were expected to have wingbeat frequencies within a limited range. Figure 4.15 presents a comparison of wingbeat frequency profiles from three continuous nights’ observations (11 – 14 February 1999) at Bourke. The wingbeat frequencies are concentrated in narrow ranges at all altitudes, which suggests a single species was predominant on each night. The frequencies, however, varied with altitude in a very similar manner for all nights, decreasing with height by 2 – 6 Hz over the IMR detection altitudes. Upper air data for 21 h at Cobar (about 160 km south of Bourke) show that temperature inversions were present below 400 m geopotential height (~135 m above ground level) and that the lapse rates above the inversion were linear and similar on all three nights (Figure 4.16). The wingbeat frequencies were slightly higher on the night of 13 – 14 February, when the temperature was about 2 °C warmer than that on the two previous nights. These results appear to confirm the earlier findings (Drake & Farrow 1983) that the wingbeat frequency is affected by temperature.

(a) 10 (b) (c) 10 2 1250-1300m 0 0 0 60 20 4 40 10 2 1100-1150m 20 0 0 0 10 40 20 950-1000m 20 5 0 0 0 60 40 10 10 800-850m 20 0 0 0 Count 50 20 20 650-700m

0 0 0 40 40 50 500-550m 20 20 0 0 0 100 100 40 350-400m 50 50 20 0 0 0 150 100 100 20 200-250m 50 50 0 0 0 20 30 40 50 20 30 40 50 20 30 40 50 11-12 Feb 1999 12-13 Feb 1999 13-14 Feb 1999 Figure 4.15 Comparison of Wingbeat Frequency Profiles from 11-14 Feb 1999 at Bourke The frequencies are estimated from targets detected by the IMR operating in stationary-beam mode during: (a) 20 – 01 h of 11 – 12 Feb 1999 (3592 targets); (b) 20 – 02 h of 12 – 13 Feb 1999 (2551 targets); and (c) 20 – 00 h of 13 – 14 Feb 1999 (1031 targets). 3 Wingbeat Frequency of an Insect Target 137

1600 11-12/02/99 (°C) 1400 12-13/02/99 (°C) 13-14/02/99 (°C) 1200

1000

800

600

Geopotential Height (m) Height Geopotential 400

200

18 20 22 24 26 28 30 Temperature (°C) at 21h from Cobar Upper Air Figure 4.16 Profiles of Temperature from Cobar Upper Air Sounding at 21h for Three Nights in February 1999 Data is from Australian Bureau of Meteorology.

The relation of wingbeat frequency and temperature has been investigated via a regression analysis. The migrating insects detected between 20 and 01 h on the night 11 – 12 February at Bourke were believed to be predominately Chortoicetes terminifera. Bourke and Cobar were under the same weather system on this night and therefore can be expected to have had a similar temperature profile. A weighted regression of wingbeat-frequency against the Cobar upper-air temperature for this night gives a=15.77, b=0.52 (20.4 – 28.4 °C, n=8, R=0.995, P<0.0001, Figure 4.17b). The temperatures are linearly interpolated to the heights of the radar range-gates from the regular radio-sonde at 21 h at Cobar. The wingbeat frequencies estimated for large insects ( > 60 mg) from rotary-beam samples show a similar relationship of a=15.11, b=0.53 (n=8, R=0.992, P<0.0001). Figure 4.17 shows a comparison of the wingbeat frequency variation with temperature for these three consecutive nights during 11 – 14 February 1999. The variation rates of wingbeat frequency with temperature were very similar. The slight differences could result from the presence of other insect species, differences in population age-structure, or variations in wingbeating that may also be affected by other physical factors, for example, humidity.

As the slopes of three fitted lines in Figure 4.17b are so close, the rate at which wingbeat frequency changes with temperature could possibly be used in species identification. A further preliminary study found that the Australian plague locust has a rate of 0.5 Hz wingbeat frequency decrease with 1 °C temperature drop. The regression

138 Chapter 4: Characterisation of IMR Targets

50 (a) 45 Mean

30 11-12/02/99 25 12-13/02/99 13-14/02/99 20 35 (b) WBF = 13.30 + 0.62 T Median Wingbeat Frequency (Hz) Wingbeat Frequency 30 WBF = 15.77 + 0.51 T

11-12/02/99 12-13/02/99 25 13-14/02/99 WBF = 11.94 + 0.66 T

20 22 24 26 28 30 32 Temperature (°C) Figure 4.17 Variation of Wingbeat Frequency with Temperature Wingbeat frequencies are plotted as mean (a) and median (b) values for each radar gate range. Wingbeat data are same as in Figure 4.13. slopes from the IMR observations in spring were found often higher than those in summer and autumn, e.g. a=2.99, b=1.07 (19.1-27.9 °C, n=8, R=0.989, P<0.0001) and a=6.81, b=0.97 (18.9-28.0 °C, Cobar, n=8, R=0.939, P<0.0001) for the 20-21 h observations at Bourke in the evening of 27 and 28 September 2000 respectively.

The temperature dependence suggests that wingbeat frequencies cannot be simply compared without consideration of environmental temperature. For example, many moth targets similar to the size of plague locust detected in winter and early spring had wingbeat frequencies of 30-35 Hz. If the temperatures were taken into account, their wingbeat frequencies would be much higher than those of locusts flying in summer at the same temperature.

3.2 Wingbeat Modulation Pattern

Wingbeating may produce a characteristic modulation pattern, which could be reflected in the harmonic content of the signals. If this is true, the intensity ratios of the harmonics could perhaps be used as another indicator of target identification (Drake et 3 Wingbeat Frequency of an Insect Target 139 al. 2002b). Figure 4.18 shows a comparison of harmonic intensities for samples of echoes from two different seasons. It seems that the third order harmonic was stronger than the second order harmonic, by about 1.5 :1, during the February 1999 observations, while they were almost equal during September 2000.

70 160 250 (a) 200-1300m (b) 200-1150m (c) 200-1300m 20-01h 140 20-02h 60 20-00h 11-12/2/99 12-13/2/99 13-14/2/99 200 1604 120 946 50 413 100 150 40 80 30 100 60 20 40 50 20 10

0 0 0

(d) 200-1400m140 (e) 200-1400m100 (f) 200-1400m 40 18-00h 19-03h 29-03h 120 Target Count Target 26-27/9/00 27-28/9/00 80 28-29/9/00 284 1155 959 30 100 60 80 20 60 40

40 10 20 20

0 0 0 -1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0 -1.0 -0.5 0.0 0.5 1.0 Amplitude Ratio of 3rd to 2nd Harmonic (logA - logA ) 3rd 2nd Figure 4.18 Distributions of Harmonic Amplitude Ratio (logarithmically transformed) for Nights in Summer and Spring Data is from the stationary-beam observations of the Bourke IMR during February 1999 (a, b, c) and September 2000 (d, e, f).

3.3 Wingbeat Frequency and Insect Size and Shape

The relationship between the size and the wingbeat frequency of each individual target can now be directly determined from the signals obtained under the rotary-beam mode observations (Wang & Drake 2004). Although wingbeat frequency can be estimated for only ~50% of the signals (see Chapter 3), the direct link with size and shape that rotary- beam observation allows will give a much clearer picture of the target characteristics than is possible from comparisons between different samples obtained from stationary- beam observations (for wingbeats) and rotary-beam observations (for RCS parameters). Extraction of wingbeat parameters from rotary-beam samples should therefore enable more accurate identifications to be achieved.

140 Chapter 4: Characterisation of IMR Targets

The difficulty of deducing the relationship of mass and wingbeat frequency from observations made separately under the rotary- and stationary-beam modes is illustrated in Figure 4.19, which shows distributions of these quantities for an occasion when two different types of insect were flying together. The smaller species was dominant at all altitudes; the number of targets under 60 mg is almost double that of the broad spread of bigger targets over 60 mg (Figure 4.19a). The number in the low-frequency peak below 35 Hz is obviously bigger than that for the high-frequency peak between 60 – 80 Hz, in both the stationary- and rotary-beam data (Figure 4.19b, c). However, it cannot be simply concluded that the smaller insects have the wingbeat frequencies under 35 Hz. This uncertainty is resolved with the use of rotary-beam mode data. Figure 4.20 shows a scatter plot (with contours added) of wingbeat frequency against mass from the same dataset as in Figure 4.19. It is very clear that the high-frequency peak between 55 and 75 Hz is due entirely to small insects (< 50 mg). The bigger insects all have frequencies under 35 Hz as also do most of the remaining small ones, although harmonics at about 20 Hz from the conical scan are evident. This also illustrates that bigger insects have more chance to have their wingbeat frequency detected than small insects, perhaps because their wingbeat modulations are stronger.

60 500 300 (a) (d) (g) 500-650m 400 40 200 Rotary Beam Mode Stationary Beam Mode 300 Rotary Beam Mode 200 100 2546 20 359 2350 100 0 0 0

600 (b) 120 (e) 800 (h) 350-500m 600 400 80

Count 400 40 200 3988 719 200 3782 0 0 0 800 (c) (f) 1000 (i) 200-350m 80 800 600 600 400 >400 40 >84 400 200 4394 842 4290 200 >84 0 0 0 0 100 200 300 400 20 30 40 50 60 70 80 20 30 40 50 60 70 80 Mass (mg) Wingbeat Frequency (Hz) Figure 4.19 Profiles of Mass and Wingbeat Frequency from the IMR Observations at Bourke on the Night of 15-16 Nov 1999 The distributions of mass (a, b, c), of wingbeat frequency from stationary-beam samples (d, e, f) and from rotary-beam samples (g, h, i) are for insects detected in three height intervals between 200 – 650 m, between 18 – 01 h. Harmonics of the 5 Hz conical-scan in the rotary-beam samples (see Chapter 3) have produced some false frequencies at about 20 Hz, and possibly some at 25 Hz. There were only a few insects detected above 650 m. 3 Wingbeat Frequency of an Insect Target 141

90 200-650m, 18-01h, 15-16 Nov 1999 80 2-600mg, 21-91Hz, 4721 targets, Bourke

40 Wingbeat Frequency (Hz) 30

100 200 300 400 500 Mass (mg) Figure 4.20 Scatter Plot of Wingbeat Frequency against Mass for the IMR Observation from Bourke on the Night of 15-16 Nov 1999 Frequencies of 20 Hz, which are predominantly from the conical-scan and its harmonics, have been excluded. As usual, the mass estimates for large insects are spread over a broad range (at the bottom of the graph). Contours indicate equal counts for the scatters (targets). 5 Hz conical-scan is presumed to produce some harmonics at 25 Hz, and possibly even 30 Hz.

However no useful relationship has been found between harmonic patterns and body shape factors to allow species discrimination.

4 Identification of Chortoicetes terminifera from IMR Echoes

The various echo characters described in the preceding sections of this chapter have been applied to the test of identifying Australian plague locusts, Chortoicetes terminifera, from IMR echoes. To determine if the plague locust was the dominant species in the airborne migrating population detected by the Bourke IMR, the APLC light-trap, field-survey and wind-trajectory data were used to analyse the locust emigration and immigration events in different regions, the synchrony of locust appearance in time and space, and suitable weather systems for transportation of locust populations between different regions. The light-trap data was related to the locust migration types (see Section 1.1 of Chapter 5) detected by the IMR when crosschecking migration events. For example, if the migration was an overflight event, the light traps located upstream and downstream according to the prevailing wind for that night were

142 Chapter 4: Characterisation of IMR Targets checked rather than the one nearest the IMR site. Also, the efficiency of light-trap for monitoring migrations was taken into account (refers to Section 1.4.1 of Chapter 1). The echo characters were identified from 25 nights when large numbers of C. terminifera were indicated migrating over the Bourke region and these characters were then used to examine the rest of IMR observations at Bourke during the three years June 1998 to May 2001 and at Thargomindah during September 1999 to January 2001 to identify possible migration nights.

The echo characters of C. terminifera that have been identified from this sample are summarised as follows:-

Mass range of 50 – 250 mg (68% probability), with a peak position about 130 mg (before scale factor applied, e.g. Figure 4.3);

aa20 peak in the range 0.6 – 0.8 with a broad spread, and aa40 peak at 0.20 – 0.25 (e.g. Figure 4.6a, b);

aa42 is greater than 0.25 (typically 0.35 – 0.40), indicating its RCS has a maximum when its longitudinal body axis is parallel to the polarisation and a secondary maximum when this axis is perpendicular (e.g. Figure 4.5a, b);

σ yyσ xx peak in the range 0.2 – 0.3, with a long right tail (e.g. Figure 4.8a, b, Figure 4.10a, b);

Wingbeat frequency between 25 – 35 Hz depending on temperature, increasing by approximately 1 Hz with every 2 °C increase in temperature (e.g. Figure 4.15, and Figure 4.16).

Usually the third harmonic is stronger than the second.

Migration events of C. terminifera over Bourke during the three years of observations according to these criteria have been identified (listed in Appendix B). As expected, locusts were detected mainly during summer. Their presence was inferred on about 140 nights, and on more than 100 nights between November and April targets with this set of characters were predominant. No locust migration was detected during May to October. As the detected movements did not occur at obvious generation intervals, it appears that asynchronous generations were commonly present. The migrations of Australian plague locust over Thargomindah, on the other hand, were not 4 Identification of Chortoicetes terminifera from IMR Echoes 143 as intense as over Bourke during the operation period less than one year, and no useful conclusions can be drawn about them.

5 Discussion

Without direct measurements of identified specimens, insect targets have been shown to be identifiable from their echo signatures acquired in both the rotary- and the stationary- beam modes of IMR observations. Identifications are based on a library of echo characters that has been built up by historical matching of IMR observations and ancillary information; a direct match with aerial samples would be ideal but not feasible during this study due to the lack of equipment and the remote locations of IMR sites. It has been found that no single echo parameter is sufficient to identify a particular species, but use of a combination of all available echo parameters permits identification with considerable confidence.

Many attempts have been made to estimate insect mass from measured RCS values, as a means of identifying target species (Russell & Wilson 1997). However, observational practice shows that this approach is only partially successful, and works best when there are only a few relatively distinct species (Aldhous 1989). Size overlapping is very common between species. Sexual dimorphism, which occurs in almost every species, often incorporates size differences, with females usually larger. There is often a large range in size among individuals of the same species, for example, 110 – 500 mg in Australian plague locusts, depending on factors such as age and nutritional condition (Hunter 1982a, 1989). In addition, different species with similar ecological niches cohabit and sometimes develop synchronously under natural conditions, and sometimes are similar in both size and behaviour. For example, oriental armyworm Mythimna separata and black cutworm Agrotis ipsilon moths, which have similar sizes as well as many other biological characters, are often caught together in a light trap (Li et al. 1964). Furthermore, estimation errors will arise from the noise in the radar hardware and from any shortcomings in the analysis algorithm, and these make the problem of size overlapping even worse. Therefore, insect targets cannot be identified adequately from size information only. However, concentrated mass

144 Chapter 4: Characterisation of IMR Targets distributions have often been seen in IMR datasets, and as target size is an obvious indicator of insect species, it remains a key identification factor.

The 3rd-order polynomial mass estimator introduced here adequately spans the 2-3 orders of magnitude range of airborne-insect sizes. Practically, it eliminates the possibility of false size-peaks and discontinuities, and simplifies the task of choosing mass estimators (Smith et al. 1993; Chapman et al. 2002a) and of dealing with targets of different shapes. It is based on the reliable laboratory measurements of Aldhous (1989), and produces results that are comparable to those from other mass estimators.

Using peak values and distribution patterns of body shape factors to map the IMR observations provides a reliable way to identify the majority of insect targets. The ratio aa42 differs significantly between the Rayleigh and Mie scattering regions (Aldhous 1989). However, Mie RCS characters can result from either target size and/or target shape. The ratios aa20, aa40 and σ yyσ xx will discriminate targets more efficiently than ratio aa42. This last quantity varies in a complex way with target size and shape, and needs to be used in conjunction with the mass distribution. The shape factor σ yyσ xx was used to help narrow target selection for particular species (Chapman et al. 2002b; Chapman et al. 2005; Chapman et al. 2006). Sometimes, one of these shape factors is enough to separate different targets; more often, several are required to achieve target identification. The patterns of shape characters for Australian plague locusts, identified from dates when this species was known to be present, have been found to be very constant. No simple relationship between shape factors and mass has been found from the IMR observations; this could reflect the diversity of insect fauna detected by the IMR. Overlapping of wingbeat frequency ranges among species presents a major difficulty for its use in target identification. It is not reliable to identify targets by simply comparing wingbeat frequencies detected at different heights or on different days, due to the dependence of frequency on temperature. For the same species, 3-4 Hz difference of wingbeat frequency at different heights or time on the same night has often been observed in the IMR data. Nevertheless, the estimates of wingbeat frequency have shown concentrated distributions, often <10 Hz wide for a night with a single predominant migrating species. Scaling the wingbeat frequencies to a standard temperature (e.g. 25 °C) should enhance the utilisation of wingbeat frequency as an 5 Discussion 145 identification factor. Further study on the relationship of wingbeat frequency and temperature for a variety of species would be required before the frequency change rate with temperature could be employed as a species identification factor.

A mixed population of insect species makes species identification difficult, if not impossible, if wingbeat frequencies are obtained only from the stationary-beam observations, as individual target sizes cannot then be determined. Matching the observations made in the rotary and stationary beam modes is always risky, as the population components, e.g. ratios of species and sex, may vary with time. In addition, the probability of wingbeat frequency being detectable probably differs between large and small insects. In such cases, wingbeat frequencies estimated from the rotary-beam observations are potentially valuable.

Target species can be identified from echo characters either by a theoretical relation or by empirical comparisons with characters observed when target identity has been established by traditional trapping and ecological surveys. An example of the former is the inverse relation of wingbeat frequency to insect size, especially wing length (Riley 1979; Riley & Reynolds 1983; Dudley 2000a). Thus, wingbeat frequency also gives a clue about the target’s size. Such incidental information can help to increase confidence about the identifications, especially when a diverse population is present. The examples of the latter are the identification of diamondback moths Plutella xylostella, carabid beetles Notiophilus biguttatus and lacewings Chrysoperla carnea from vertical-looking radar echoes by combination of aerial sampling and laboratory measurements of these captures on mass, size and radar echo signatures, through the single shape factor σ yyσ xx might be lack of power to separate the species from a much diverse airborne fauna (Chapman et al. 2002b; Chapman et al. 2005; Chapman et al. 2006).

5 Tracing Migration Courses of Australian Plague Locust

1 CHARACTERISATION OF PLAGUE LOCUST FLIGHT...... 148 1.1 IDENTIFICATION OF NOCTURNAL FLIGHT ACTIVITY ...... 148 1.1.1 Local dusk trial flight...... 149 1.1.2 Emigration ...... 150 1.1.3 Overflight ...... 150 1.1.4 Immigration...... 150 1.1.5 Emigration and overflight...... 151 1.1.6 Emigration and immigration...... 151 1.1.7 Overflight and immigration...... 152 1.1.8 Emigration, overflight and immigration ...... 152 1.2 CHARACTERISTICS OF PLAGUE LOCUST MIGRATION ...... 153 1.2.1 Dusk take-off ...... 153 1.2.2 Migration duration and height...... 155 1.2.3 Migration displacement and orientation direction...... 156 1.2.4 Landing ...... 157 1.3 WEATHER SYSTEMS ASSOCIATED WITH PLAGUE LOCUST MIGRATIONS...... 157 2 ESTIMATION OF PLAGUE LOCUST MIGRATIONS ...... 159 2.1 TRAJECTORY ANALYSIS METHOD ...... 159 2.2 ESTIMATING MIGRATION PATHS OF PLAGUE LOCUST FROM IMR OBSERVATIONS...... 161 2.2.1 Major locust migrations detected in 1999 – 2000...... 162 2.2.2 Seasonal patterns of locust movements detected in 1998 – 2001...... 172 3 DISCUSSION...... 176

The potential value of entomological radars arises largely from their ability to detect movements of migratory insects directly. Although entomological radars have been used to study daytime flying insects including C. terminifera (Reid et al. 1979), they are especially valuable at night when other means are not capable of detecting migrating insects without disturbing them (Reynolds 1988; Zhai 1999). The IMRs in inland eastern Australia provide information on nocturnal insect activities in this remote region in near real-time (Drake et al. 2001; Drake et al. 2002a). Automated data processing of the observations, combined with on-line publishing of the result, potentially enables near real-time delivery of information about the identities and movements of migrating insect populations. An objective of the network was that migrations of key migratory insect pests between the arid inland rangeland and the main agricultural regions of

147 148 Chapter 5: Tracing Migration Courses of Australian Plague Locust eastern Australia can be monitored reliably, so that forecasting and management of these pests can be undertaken more efficiently and effectively.

To evaluate the application of IMR network, migration events (comprising about 140 nights, see Appendix B) of Australian plague locust Chortoicetes terminifera, identified from the characteristics of IMR echoes (see Chapter 4) obtained during in the period June 1998 – May 2001 at Bourke and 31 nights between September 1999 – May 2000 at Thargomindah, are examined with ancillary information, such as upper air data from Bureau of Meteorology station at Cobar (the closest station to Bourke), grid maps of rain events, 6-hourly pressure charts at mean sea level (between December 1999 and May 2001), fortnightly NOAA NDVI satellite images, light-trap catches and field- survey data from the APLC. The nocturnal flight activities of C. terminifera are classified from the IMR observations and the migration characteristics analysed in relation to weather conditions and patterns. Target trajectory parameters estimated from the IMR observations are used to estimate migration pathways, from which likely source and destination areas are located. The locust occurrences are thus identified in this way and validated with relevant ancillary data. Finally, the development and distribution of the plague locust in the eastern Australia in the three seasons is reviewed with the evidence of locust movement as detected by the IMR network.

1 Characterisation of Plague Locust Flight

The major nighttime flights of C. terminifera detected at Bourke IMR between 1998 and 2001 are examined in this section to categorising their activities and characterising the migration process.

1.1 Identification of Nocturnal Flight Activity

The time-series of vertical-profile data from the IMR observations indicates the flight types and possible displacement ranges of the flying insects. The nights with dominant C. terminifera flights are categorised according to the profile patterns obtained by the IMR, and form the sample for the analysis of migration in the remaining part of this 1 Characterisation of Plague Locust Flight 149

Chapter. Examples of typical patterns are presented here, along with comments on their effects on the local population.

1.1.1 Local dusk trial flight A dusk flight stimulated mainly by decrease of light intensity occurs in many insect species (Dingle 1996). Flight in short time, low altitudes, and short distance, usually starting just after sunset, is the typical characteristic of trial flight in feeding and mating among many species. It might be an attempt to emigrate that is unsuccessful due to unfavourable weather conditions. However, there is no way to confirm this from IMR observations. The local population, i.e. the population within a few ten kilometres of the IMR surrounding area, is expected to remain largely unchanged on both population size and structure. Plague locusts have occasionally been detected flying by the IMR data for only a short time, approximately 1 hour, after sunset (Figure 5.1a).

Figure 5.1 Time-series of vertical-profile Showing (a) Local Dusk Flight and (b) Emigration (a) on 26 March 2001, a short-time peak of flight activity was detected after sunset that was at 18:20 h; surface temperature was 20.7 °C and wind speed was 0.5 m/s at 19:00 h. (b) On 25 February 2000, targets were detected increasingly at high altitudes during the 4 h following sunset (at 18:54 h) indicating that the locusts were emigrating from a large area nearby. Surface temperature was 29.4 °C and wind speed was 1.1 m/s at 19:00 h. Sunset time is calculated using the on-line calculator of the Astronomical Applications Department of the US Naval Observatory (http://www.aa.usno.navy.mil).

150 Chapter 5: Tracing Migration Courses of Australian Plague Locust

1.1.2 Emigration Contrasting with trial flight, an emigration event is taken to be characterised by a massive flight of locusts which is observed to commence after sunset and to persist over several hours at higher altitudes. Such flights must lead to a proportion of the locust population moving out substantial distances from the nearby area (Figure 5.1b).

1.1.3 Overflight An event is classified as an overflight when plague locusts are detected by the IMR at higher altitudes after the time when the normal dusk peak of flight has ceased. This pattern indicates that the locusts are passing over the IMR site and originate from an area some distance away (Figure 5.2a). The local population of adult locusts in the vicinity of the IMR may remain unchanged, if there is little take-off and landing.

Figure 5.2 Time-series of vertical-profile Showing (a) Overflight and (b) Immigration (a) On 19 February 1999, the dusk flight activity was small, but a large number of locusts were detected around midnight, indicating the source area was some distance from the IMR. Sunset was at 19:01 h. (b) On 02 February 2001, a tropical trough was approaching the IMR location. Rain was detected during 19- 20 h, surface temperature was 24.0 ~ 24.5 °C and the wind was easterly at 1.5 ~ 2.5 m/s during the following 6 hr.

1.1.4 Immigration Locusts arriving from some distance external to the IMR produce a late peak, in which the targets typically appear at some height first and then later occur more at lower levels 1 Characterisation of Plague Locust Flight 151

(Figure 5.1b). The time difference between the later peak and the usual take-off peak just after sunset indicates the time of the immigrating population has travelled before reaching the IMR site. This is based on the assumption/fact that the acridoids do not take off later at night (see Section 2.3.3 of Chapter 1). Immigration often appears as a sudden increase of adult locusts or ages differing from those of the local population.

1.1.5 Emigration and overflight The vertical-profile time series shown in Figure 5.3a is typical when an emigration event occurs over a large area around the IMR site and the up-wind region. On this occasion, the IMR detected an intense local take-off at sunset and then a continuing overflight indicating a long-distance migrating movement from the up-wind area. The adult density often decreases suddenly over a wide area after such an emigration.

Figure 5.3 Time-series of vertical-profile Showing (a) Emigration and Overflight and (b) Emigration and Immigration (a) On 18 February 1999, surface temperature was 23.9 °C and wind speed was 2.6 m/s at 19:00 h; sunset was at 19:02 h. (b) On 8 March 2001, surface temperature was 30.4 °C and wind speed was 0.9 m/s at 19:00 h; sunset was at 18:43 h.

1.1.6 Emigration and immigration An exchange of locust populations occurs when a local population moves out and other populations move in from elsewhere (Figure 5.3b). The local population density depends on the ratio of emigrants and immigrants. Such an event may be hardly

152 Chapter 5: Tracing Migration Courses of Australian Plague Locust detectable in terms of the local population if the inputs and outputs are similarly sized and if the age structures are not much different.

1.1.7 Overflight and immigration Occasionally it would be observed that a dusk take-off peak was absent but there was evidence of an overflight and later of flight at low altitudes (Figure 5.4a). This indicates that locusts took off from a source area at some distance from the IMR, and that some of them landed in the vicinity of the IMR. In these circumstances, the local population will generally increase.

Figure 5.4 Time-series of vertical-profile Showing (a) Overflight and immigration and (b) Emigration, Overflight and Immigration (a) On 22 February 1999, surface temperature was 30.0 °C and wind speed was 0.8 m/s at 19:00 h; sunset was at 18:57 h. (b) On 10 February 2001, surface temperature was 32.1 °C and wind speed was 2.4 m/s at 19:00 h; sunset was at 19:09 h.

1.1.8 Emigration, overflight and immigration At times, the observed pattern of echo occurrence indicates that emigration, overflight, and immigration have all occurred in one single night (Figure 5.4b). The net effect on the local population level depends on the ratio of emigrants and immigrants, and the local population change may not be identified if exchanged populations were not very different in terms of overall numbers and age. 1 Characterisation of Plague Locust Flight 153

Although the nights of C. terminifera flight have been classified as emigration, overflight, and immigration events according to the major patterns of the time-series of vertical-profile plots (see Appendix B), there are no distinctive periods of emigration, overflight and immigration in the 140 nights of detected locust migrations. Migrating locusts were often observed by the IMR over several consecutive nights in a row, for example during 1 – 3 Jan 2000 and 1 – 6 March 2001, indicating that this is not an uncommon plague locust long-distance migration pattern. In addition, migration periods are difficult to categorise into generations due to common asynchronous generations in a wide area and absence of the IMR observation occasionally due to mechanical failure or maintenance service within possible massive migration waves. Nevertheless, the time-series of vertical-profile provides the information for the insect activities and could be used to trace the migrations.

1.2 Characteristics of Plague Locust Migration

In this section the principal features of C. terminifera migration evident from the IMR observations are related to existing findings of the long-distance migration of this species.

1.2.1 Dusk take-off The dusk take-off peaks of C. terminifera are strongly associated with environmental conditions. The starting time of dusk take-off, determined as the time when detected targets increase rapidly, was 16.5±16.8 min (-18 – 94 min) after sunset (Figure 5.5a). The time delay between taking-off from the ground and being detected by the IMR due to the radar blind range, i.e. the time for a locust to reach first gate height, is considered short, possibly within a few minutes (Riley & Reynolds 1990). Surface air temperature was 29.0±3.7 °C (20.8 – 35.9 °C) (Figure 5.5b), and wind speed was 0.8±0.8 m/s (0 – 3.6 m/s) (Figure 5.5c), at the beginnings of dusk take-off. The peak time and duration of dusk take-off are difficult to determine, because of the possible flights from nearby area.

The dusk take-off of C. terminifera detected by the IMR appeared to be very similar to that observed by Clark (1971). The starting times of take-off were clustered shortly after sunset, 70% falling within the 30 min following sunset, indicating that the essential stimulus is decreasing light intensity. The recorded minimum temperature for

154 Chapter 5: Tracing Migration Courses of Australian Plague Locust

(a) 15 Time after Sunset (min) 10

0 -20 0 20 40 60 80 100

10 (b) Surface Temperature (°C)

Number of Nights of Number 20 22 24 26 28 30 32 34 36 20 (c) 15 Surface Wind (m/s) 10

0 0.0 0.4 0.8 1.2 1.6 2.0 2.4 2.8 3.2 3.6 4.0

Figure 5.5 Conditions of C. terminifera Take-off Histograms of (a) starting times for locust dusk take-offs detected at Bourke on 95 nights during June 1998 – May 2001, (b) temperatures, and (c) wind speeds for the starting times of take-off from the automatic weather station incorporated with the IMR. dusk take-off was in very good agreement to the 20 °C obtained from field observations (Clark 1969, 1971). Temperature is usually not a limiting factor in summer in this region, for example, the surface air temperature at 19 h was 29.9±4.6 °C over the three summers (December-February, 1998 – 2001) from the AWS incorporated with the IMR. It is almost certain that migrations of C. terminifera were related to disturbed weather (see Section 1.3), though there were a few occasions when the dusk take-off occurred in almost calm conditions, which has previously been recorded only in a high density population (Clark 1971). The recorded calm conditions may attribute to the location of the AWS which is located in a not ideal area surrounded by a big shed (in about 20 m to the west) and some eucalyptus trees around. The AWS location is not ideal due to the installation requirement within the airport. Examination of the weather patterns revealed that evenings immediately preceding the arrival of disturbed weather systems, most often a tropical trough were associated with high levels of locust activity on the IMR. In addition, the most intense take-offs occurred in light winds, often with wind shift. On the other hand, there were a few calm or almost calm evenings with no take-off, nor a delayed peak, when the previous and the following evenings had a normal dusk take-off soon after sunset, which suggests that adult locusts were ready to emigrate but inhibited 1 Characterisation of Plague Locust Flight 155 by unfavourable weather conditions at dusk. The maximum wind speed observed by the AWS at the IMR site during take-off period did not reach the recorded upper limit of 10 m/s, which though is reported even not suppressing the take-off on one occasion (Hunter 1982b). The IMR observations therefore support that the view that surface light wind at dusk is a critical factor promoting initiation of C. terminifera migration as reported by earlier workers (Clark 1971; Hunter 1981b).

1.2.2 Migration duration and height The migrations of C. terminifera detected by the IMR at Bourke often lasted for 6 – 8 hr (7.4±1.7 hr for 134 nights, Figure 5.6a), starting after sunset and ceasing shortly after midnight. Shorter duration migration was sometimes detected, of approximately 4-6 hr duration. The migration distances, however, often cover no more than 200 km in one night (see page 162).

30 (a) (b) (c) 20 25 15

15 20 10 15 10

10 5 5

Number of Nights of Number 5

0 0 0 46810 800 1000 1200 1400 12 14 16 18 20 22 24 26 28 Duration (h) Ceiling (m) Temperature (°C) Figure 5.6 Histograms for (a) Duration and (b) Layer Ceiling of C. terminifera Migration at Bourke and (c) Temperature of Ceiling Height at Cobar The Migrations of plague locust detected at Bourke are examined for flight duration (a) and upper boundary at 21 h (b), when the temperature (c) at the same altitude is matched from Cobar radiosonde.

Concentration of the migrating locusts in a layer was usually seen in the IMR observations. The lower layer boundary was often clear, at about 300 m agl, while the upper boundary was usually not well defined, but mostly above 800 m and below 1200 m agl (Figure 5.6b). Occasionally the formation of two layers at different heights was detected. Temperatures at the upper boundary (Figure 5.6c) were often about 20 °C in summer (21.3±3.3 °C for 84 nights in December-February). The minimum temperature at the upper boundary was 12.7 °C at 1100 m on one occasion.

The detected migrations of C. terminifera by the IMR at Thargomindah, however, were mostly continuous through the whole night and lasted for 9–10 hr. In

156 Chapter 5: Tracing Migration Courses of Australian Plague Locust addition, migrating locusts were often present throughout the whole range of observation heights. The duration of locust migration, therefore, is most likely affected by temperature, as temperatures at Thargomindah are generally higher than at Bourke.

1.2.3 Migration displacement and orientation direction

The nocturnal migrations of C. terminifera had always been seen to have a common displacement direction and to showed collective orientation from the IMR observations. Usually a crab angle, i.e. the smaller angle between the displacement direction and the body alignment, occurred for each individual flying locust. Figure 5.7 shows the distributions of nightly averaged displacements and crab angles demonstrated by the migrating locusts over Bourke from the three seasons of the study period. The displacements were clustered with a mean value of 296°. However, the displacement directions for the migrating locusts detected by the Thargomindah IMR (see Figure 5.23) were not so concentrated and had no obvious preferential direction. In addition, no unique crab angle has been found, through its distribution is not normal with a 0 mean, i.e. skewed (discussed in Section 2.4.3 of Chapter 6). Neither the displacement nor the orientation direction pointed to particular azimuths. Chapter 6 discusses the orientation behaviours of C. terminifera in more detail.

0 (a) (b) 20 108 nights in total

296 15

270 90 10 Percentage (%) Percentage 5

0 180 -90 -60 -30 0 30 60 90 Displacement (degree) Crab Angle (degree) Figure 5.7 Histograms of Nightly Displacements (a) and Crab Angles (b) of Migrating C. terminifera over Bourke during 1998-2001 The displacements and orientation directions for all good-quality echoes in the radar detection range are averaged nightly over all locust migration nights. The crab angle is calculated by subtraction of displacement direction from the body alignment. Nights with wind shift/shear during the detected peak time of plague locust movements are excluded and only those nights with obvious preferential directions are included (P < 0.10 for the Rayleigh test). 1 Characterisation of Plague Locust Flight 157

1.2.4 Landing It is difficult to determine whether the descent and landing behaviour of plague locusts is regulated by endocrine agents, or is mainly stimulated by unfavourable weather conditions, in part because the required radio-sonde measurements of atmospheric conditions are not available. However, it seemed that the termination of migration coincides with the disappearance of the nocturnal temperature inversion. Locust migrations often ceased shortly after midnight at Bourke, where the nocturnal temperature inversion often develops in the evening and temperature at the inversion ceiling is usually above 20 °C (from Cobar radio-sonde data), but not at Thargomindah where (subtropical) temperature is much higher and locust migrations usually lasted the whole night (see above). Therefore, warmer temperatures and stronger winds could be favourable physical factors promoting locust migration.

1.3 Weather Systems Associated with Plague Locust Migrations

In southeastern Australia migrations of C. terminifera usually occur between late spring (November) and mid-autumn (April). During this period, the weather is usually characterised by a series of high pressure systems, often separated by a trough or a low pressure system, moving eastwards along the latitude at about 40° S (see page 36). Migration events of C. terminifera detected by the IMR at Bourke were strongly related to disturbed weather, in the form of a tropical trough or a col, both typical systems producing disturbed weather in southeastern Australia in summer (Figure 5.8). In the 72 nights of C. terminifera migration between December 1999 and May 2001 (when weather maps were available), southward or southwestward movements (43 out of 72 nights) were frequently associated with warm north-easterly air flow, often on the eastern side of a tropical trough (Figure 5.8a, c). Such situations are often associated with thunderstorms (Clark 1969; Farrow 1979). However, southerly or southwesterly migration was also observed on the eastern side of a cold front associated with a temperate low pressure system, on the western flank of the semi-permanent Pacific high (the blocking Tasman Sea high) (Figure 5.8e). In contrast, northward movements (29 out of 72 nights) were normally in the cooler southerly air flow on the western side of a tropical trough (Figure 5.8b, d). Occasionally the movements were in the highly modified cool southerlies or southwesterlies, on the western side of a cold front

158 Chapter 5: Tracing Migration Courses of Australian Plague Locust associated with a mid-latitude low pressure system and the eastern ridge of the Indian Ocean high (Figure 5.8f) (Reid et al. 1979; Drake & Farrow 1983). Tropical troughs were the predominant systems within which the plague locusts migrated. No migration of C. terminifera was detected in settled weather with light winds under anticyclonic conditions (i.e. under a high pressure system) at Bourke. However, emigrations, overflights, and immigrations cannot be connected to particular parts of the weather systems. Some examples of intense C. terminifera migration in a variety of synoptic situations are described in the following section.

Figure 5.8 C. terminifera Migrations Associated Typical Summer Weather Patterns Mean sea level (MSL) pressure charts for 22:00 h on (a) 24 February 2000, (b) 28 February 2000, (c) 12 February 2001, (d) 27 February 2001, (e) 03 February 2001, (f) 03 December 1999 (from the web site of the Australian Bureau of Meteorology). Australian plague locusts, detected by the IMR at Bourke, migrated southwestward on (a), (c) and (e); northwestward on (b), (d) and (f). 2 Estimation of Plague Locust Migrations 159

2 Estimation of Plague Locust Migrations

From IMR observations, a detected migration of plague locust can be identified into the processes described above. The approximate migration path can then be estimated, and source and destination regions located, using the track parameters estimated from the echoes by trajectory analysis (see Chapter 3).

2.1 Trajectory Analysis Method

Using upper-air wind data to estimate flight trajectory is often adopted in the study of insect migration. Downwind displacement provides the possibility of using an air parcel to simulate the trajectory of an insect migrating in the wind field. Upper winds are often uniform over a large area, and the insects flying within the same part of the weather system, e.g. on the eastern side of a tropical trough or before a cold front, are expected to move along similar pathways if there are no significant terrain changes. With the evidence of migration, wind trajectories can be calculated from upper winds and the likely origin and invasion regions of the migrants located, given the take-off time, flight duration and height are known. Wind trajectory analysis has produced very good estimates of micro-insect migrations, e.g. the migrations of brown planthopper Nilarparvata lugens in China (Rosenberg & Magor 1987; Riley et al. 1994). However, trajectory models can also generate significant errors in estimating macro-insect migrations, e.g. for corn earworm moths Helicoverpa zea (Westbrook et al. 1995a; Westbrook et al. 1995b). The large error in the trajectory estimate may arise due to the greater airspeed and possible crosswind orientation of macro-migrants, and faster wind speeds at the flight height, e.g. within the low-level jet and the nocturnal temperature inversion layer. Plague locusts are often observed by the IMR in uniform displacement with a collective orientation at ground speeds often faster by 3 – 4 m/s than the simulated wind speed by the Bureau of Meteorology Local Analysis and Prediction System (LAPS) model (Puri et al. 1998) (see Chapter 6). Therefore, wind trajectories estimated from the LAPS model may not provide accurate estimates for plague locust migrations, given that migrants fly actively during migration.

160 Chapter 5: Tracing Migration Courses of Australian Plague Locust

Migration trajectories of plague locusts could be better estimated from IMR observations. With IMR observations, displacement directions and ground speeds of individual targets are measured directly. The observed trajectory portion of individual flying insects by the IMR is shorter than that detected by scanning radar, the former is only a few seconds but the latter can last for minutes. However, the track vectors determined from the IMR are much more accurate than those measured on the PPI display of scanning radar, which may produce large error from distorted display. Assuming migrants moving through the region change their directions and speeds in the same manner as those passing over the IMR, the time-series of track vectors calculated from different individuals detected by the IMR may represent an individual movement over a large distance. Based on this assumption, the migration trajectory of a plague locust population is therefore estimated, i.e. using the flight parameters measured from different individual targets as the estimates for a single individual moving over different locations. Two types of trajectory can thus be calculated from IMR observations; a backward trajectory can be calculated to locate the possible source region whilst a forward trajectory can be used to suggest the invasion area.

A program FiTraj has been written to calculate hourly averaged track vectors from flying insects detected by the IMR and to aggregate these to form a progressive migration trajectory. The take-off time is assumed from the time of sunset, if local take- off is not observed. Peak take-off is assumed to be 30 min after sunset, when the emigrants are also expected to ascend to their transportation heights. The flight altitude and duration are determined from a time-series of the vertical-profile. Dependent on the type of trajectory, the start time is either the take-off time for a forward trajectory or the time when the target numbers decrease rapidly with height and time for a backward trajectory. Azimuthal directions and distances are thus calculated using a 1-hr time step and converted to latitude and longitude values using a scale appropriate for the IMR location. No attempt has been made to calculate backward trajectories through to the take-off region if no targets were detected in a period following sunset, and forward trajectories are terminated on the first occasion when no targets were detected at the flight height. Temperature limitation has not been considered, as the plague locusts migrate often at high temperature, and on the other hand there is no easy way to estimate upper-air temperatures either from the IMR or from the surface weather observations. 2 Estimation of Plague Locust Migrations 161

2.2 Estimating Migration Paths of Plague Locust from IMR Observations

During the three years from 1998 to 2001, Australian plague locust populations in eastern Australia went through a full plague cycle, from build-up, through outbreak and finally collapse (see Appendix C). Key population movements between different regions, therefore, will help understand the course of population distribution and development. Figure 5.9 shows suitable habitats for this species in eastern Australia and the district names used thereafter (map from APLC). In this section, the sequence of locust movements detected by the IMRs in the season of 1999 – 2000 is examined in detail in relation to ancillary evidence such as the APLC light-trap and locust survey

Figure 5.9 Habitat Map of Chortoicetes terminifera in eastern Australia Suitable habitats are shown as green shading. The districts are those used by the Australian Bureau of Meteorology. Map source: Australian Plague Locust Commission.

162 Chapter 5: Tracing Migration Courses of Australian Plague Locust data and weather charts. With the measured flight parameters from IMR observations, the migration trajectories of plague locust are calculated and their sources and destinations are located. In addition, the annual trends of plague locust movement are summarised from the observations of two IMRs during the three seasons. The scenario of metapopulation persistence is reconstructed with the evidence of locust movement detected by the IMRs.

2.2.1 Major locust migrations detected in 1999 – 2000 A series of movements between the agricultural belt of NSW and the arid interior (the Channel Country of Qld and the Far North of SA) were detected by the IMRs during the 1999-2000 plague of C. terminifera in eastern Australia. The migration trajectories are estimated from IMR observations as follows.

Early summer 1999 The first major movement from the agricultural region to the arid inland was detected by the IMRs on the night of 03 – 04 December 1999. Both IMRs detected more than 25,000 targets during the 11 hr observation. From the records of the AWS at Bourke, it was calm at sunset (19:04 h) on 3 December and for the following 2 hr, during which the temperature decreased from 29.5 to 26.5 °C (Figure 5.10d). The IMR at Bourke was almost clear of targets in the first 3 hr after sunset (Figure 5.10c). The intense migration of plague locust from the southeast appeared suddenly at about 22 h, apparently in association with an increase in temperature and winds about 10 min earlier. The overflying locusts were detected in the whole radar detection range in the following 7 hr, almost to sunrise (05:07 h) when the ground temperature was about 20.6 °C. At Thargomindah, migrating locusts were mostly below 650 m agl and the target number increased markedly after midnight (Figure 5.10a). During 22 – 23 h the wind was weak and temporarily shifted to northerly other than the prevailing southeasterly on this night (Figure 5.10b). On the Bureau of Meteorology Mean Sea Level Pressure (MSLP) weather chart (Figure 5.11a) at 22 h, Bourke was just on the west side of a pre-frontal trough line between the Pacific and Indian Ocean highs. There was an associated cold front to the south of the Australian continent. Thargomindah seemed to be on the trough line at this time. The surface temperature and wind changes at the two IMR sites demonstrated the influence from this tropical trough. The 2 Estimation of Plague Locust Migrations 163 radiosonde at Cobar at 21 h showed a southerly wind of 3 – 10 m/s and temperatures of 19 – 25 °C in the radar detection range without any obvious inversion.

Figure 5.10 Time-series of Vertical Profile and Ground Weather on the Night of 03-04 Dec 1999 Migrating adults of C. terminifera were detected by both IMRs, mostly at 200 – 1000 m agl between 22 – 04 h at Bourke and at 175 – 625 m agl between 23 – 05 h at Thargomindah.

With the flight parameters obtained by the IMRs, the trajectories of this migrating population are estimated as shown in Figure 5.11c. Assuming the plague locusts started to take off after sunset and ascended to flight heights within a half hour, which were indicated from the IMR detection at Thargomindah, the plague locusts detected at 22 h at Bourke would have flown at least 50 km already, and reached Thargomindah region by next morning if they continued their flight in the following 7 hr like other locusts detected at Bourke. The later arrivals by early next morning would have come from 170 – 280 km southeast, a source possibly up to the Nyngan- Warren region. Taking the slowly moving trough (Figure 5.11a, b) into account, the

164 Chapter 5: Tracing Migration Courses of Australian Plague Locust source region could be as far east as the Coonamble-Coonabarabran region. The observations at Thargomindah indicated that the majority of locusts came from the southeast at least 6-hr journey away, and some locusts could have reached the Channel Country of Queensland.

(c)

Figure 5.11 MSLP and Trajectories of Chortoicetes terminifera for the Night of 03-04 Dec 1999 The MSLP weather charts at 10 pm (a) and 04 am (b) are from the Australian Bureau of Meteorology website. Forward (blue solid) and backward (red dash) trajectories (c) are calculated at 150-m height intervals (1 – 3 from low to high level) using 1-hr time step. At Bourke, the 7-hr forward trajectories are calculated from 22 h and the 10-hr backward trajectories are from 04 h. At Thargomindah, 7-hr trajectories are calculated between 19 h and 02 h for possible shortest trajectories.

Similar north-westward migrations of plague locust were detected at Thargomindah in the following two nights, and the nightly detected targets were doubled in comparison with the first night. The weather patterns on these two nights remained similar to the night of 03 – 04; both IMRs were on the western side of the tropical trough which had penetrated further southeast as the Tasmania high was almost stationary. Since three consecutive nights of similar migrations had been observed at Thargomindah, it would be expected that there had been similar migrations at Bourke. Unfortunately, the IMR at Bourke was out of order between 04 and 06 December 1999.

This migrating population of plague locust must have reached the Upper Western (Far Northwest and Upper Darling) of NSW and the Far Southwest and Lower Western regions of Queensland (the Channel Country). It is virtually certain that the 2 Estimation of Plague Locust Migrations 165 population was the residual of the spring generation in the Central West Plains of NSW, where a large population had developed after good winter and spring rainfalls from overwintering eggs that required the aerial control of bands and swarms (in a total area of 43.7 km2) by the APLC in Tottenham-Coonamble region between mid October and mid November 1999 (APLC Annual Report). The light-trap at Thargomindah first caught 4 and then 2 adult plague locusts on the nights of 04 – 05 and 05 – 06 December 1999 respectively. The APLC conducted intensive surveys in the Queensland Channel Country in early November and found very low numbers of adults but no nymphs. However, two low density swarms and a large area of numerous to concentration adults were seen in the Queensland Channel Country in early December 1999. These locust adults may be from the migrants detected by the IMRs. A total 233 km2 of bands and swarms were subsequently treated by the APLC in the Windorah-Durham Downs- Nockatunga-Eromanga region from mid January to early February 2000, indicating a large subsequent generation had developed from December onwards.

Mid-summer 2000 A similar intense movement of plague locust was detected by both IMRs a month later, on the night of 02 – 03 January 2000. Figure 5.12a shows the vertical

Figure 5.12 Vertical Profile and Weather at Bourke on the Night of 02-03 Jan 2000 Time-series of target numbers (a) was from Bourke IMR. MSLP weather charts were for 10pm (b) and 04am (c) of the night, indicating the lows and troughs on the continent.

166 Chapter 5: Tracing Migration Courses of Australian Plague Locust profile and indicates a large population moved into the Bourke and surrounding area. The locusts might have emigrated from a wide area, but the large population detected by the IMR at Bourke about midnight might have come mainly from Coonamble and Coonabarabran region (the Central West Plains and Slopes of NSW), moving 200 – 250 km northwestwards in a 6-hr flight. The existing population, prior to this night, around the Bourke region thus might have flown to Thurloo Downs (the Far Northwest of NSW) within similar flight duration and direction. The locust migrants detected at Thargomindah were through the whole of this night, and apparently originated from Cunnamulla (the Warrego), 200 – 250 km to the east, and may have flown westwards as fas as Orientos (the Far Southwest of Queensland). Figure 5.13 shows the trajectories estimated from the observations of both IMRs.

Figure 5.13 Trajectories of Chortoicetes terminifera for the Night of 02-03 Jan 2000 Trajectories are shown from the intense migration detected by both IMRs below 1250 m agl, continuing for 6 hr at Bourke and for 10 hr at Thargomindah after sunset.

APLC Light-trap catches also provide additional inferential data supporting the migration. On the night of 02 – 03 January 2000, one plague locust was caught at Fowler’s Gap, NSW, 72 at Thargomindah with another 18 caught on the following night, and 9 at Birdsville, Queensland, with another 8 on the next night. From the weather charts (Figure 5.12b, c), both Thargomindah and Bourke were under the influence of a very strong tropical trough which extended south to the Australian Bight. In addition, both IMRs detected continuous locust migrations in the following two 2 Estimation of Plague Locust Migrations 167 nights when both sites were under a col. Further relocations of populations into the arid interior occurred during two nights of 11 – 13 January 2000 when both IMR sites were under the influence of a weak tropical trough in the wake of the Tasman High. These movements extended the distribution of C. terminifera northward, into southwestern Queensland (the Lower Western and Far Southwest), and indicated that the plague locusts had continuously accumulated into the arid inland area.

Later summer to early autumn 2000 Additional population movements into the arid interior occurred from late January to early March 2000. The IMR at Bourke detected westward movements starting on the night of 25 – 26 January, when the light-trap at Birdsville caught 402 plague locusts in comparison to a catch of 3 individuals on the previous night. Meanwhile, the other light-traps in this region did not report any locust catches. Southwestward locust movements might have occurred in early February based on the weather charts and light-trap data. The light-trap at Birdsville caught large numbers of plague locusts between 30 January and 11 February 2000 when the Bourke IMR was out of order. During late February to early March, heavy westward movements were

Figure 5.14 Time-series of Vertical Profile and Ground Weather at Bourke on the Night of 24- 25 Feb 2000 Emigration from Bourke region was evident from the vertical profile.

168 Chapter 5: Tracing Migration Courses of Australian Plague Locust detected continuously by both IMRs. Figure 5.14 shows an example, the night of 24–25 February 2000; an almost identical migration occurred the next night and similar but much heavier migrations were detected through the late February until the beginning of March. The light-traps at Oodnadatta and Dulkaninna in SA, Fowler’s Gap and White Cliffs in NSW, and Thargomindah caught large numbers of plague locusts during this period, e.g. 6000 at Dulkaninna on the night of 27 – 28 February 2000. The light-trap at Birdsville, however, only caught small numbers.

The night of 24 – 25 February 2000 provides an example of an intense emigration event that appeared to result in a large number or population of plague locusts redistributing into the Channel Country. At Bourke, the dusk take-off started at 19 h, 5 min after sunset, with a surface temperature of 29 °C and an east-north-easterly (ENE) surface wind of just over 1.0 m/s (Figure 5.14b). The locusts reached an altitude of 1250 m agl one hour later and mostly remained below this level over the next four hours. There was no obvious layer concentration, except slightly fewer targets were detected in the lowest radar gate range (200-350 m agl). On MSLP weather charts (Figure 5.15), Bourke was on the eastern side of a tropical trough, which penetrated from the strong Pilbara heat low on the northwest of Australia into the almost stationary Pacific high. The development of the western part of the Pacific high slowed down the decay of the trough. The almost static trough provided the similar prevailing winds for the heavy migrations in the late February. Thargomindah was under the same weather system as Bourke, and the IMR there detected a similar migration pattern of plague locust except that the migration duration was much longer. Radiosonde data from Cobar (21 h) indicated a temperature inversion near the surface, and a north-easterly wind of about 5-7 m/s below 1000 m.

Figure 5.15 MSLP Weather Charts at 10pm and 04am on the Night of 24-25 Feb 2000 A strong Pilbara heat low controls almost the whole continent. 2 Estimation of Plague Locust Migrations 169

The flight trajectories estimated for the night of 24 – 25 February 2000 are shown in Figure 5.16 from the observations of both IMRs. At Bourke, the flight was to the west, with the majority of flying locusts disappearing from the IMR shortly after midnight. A mean ground speed of 10.6 m/s for migrating locusts was measured using the IMR data for the period. It is likely that the emigrants from the Bourke region had flown westwards to the Myrnong-Monolon region in Far Northwest NSW, whilst the later arrivals may had come from the Rowena-Walgett region to the east, approximately 200-230 km each way. At Thargomindah, the source of the migrants was probably the Roma-Morven region 370-430 km to the northeast, and to the Cameron Corner- Moolawatana region at a similar distance to the southwest.

Figure 5.16 Trajectories of Chortoicetes terminifera for the Night of 24-25 Feb 2000 Migrating locusts were detected under 1250 m agl at Bourke for 6 hr after sunset and though the whole night at Thargomindah.

The above movements redistributed the populations from southwestern Queensland and central NSW into the Far North of SA and the Far Northwest of NSW, i.e. Marree-Hawker-Broken Hill-Tibooburra region. Subsequent heavy summer rain fell in this inland area would have enabled the survival and reproduction. Subsequently a significant locust infestation developed in the region of Tibooburra-White Cliffs- Wanaaring of NSW extending west to Marree of SA, requiring the APLC to control 843 km2 of bands between 25 March and 3 April 2000. The population in the agricultural belt of central western NSW decreased except in the Riverina and the South

170 Chapter 5: Tracing Migration Courses of Australian Plague Locust

West Plains regions, where adults seemed to persist rather than migrate. Between 26 February to 3 March 2000, the APLC sprayed 45 km2 of swarms in Griffith area of NSW.

Mid-autumn 2000 No massive locust movement was detected by either IMR between early March and early April. Following APLC band control in late March-early April in the inland 80 km2 of swarms were also sprayed between 8 and 15 April 2000. This indicated that the population of this autumn generation had been built up rapidly mainly from the previous immigrants in the interior after a good summer rain.

A subsequent return migration from the inland to the south eastern agricultural region was detected by both IMR units. At Thargomindah, southeastward movements of plague locust were detected on two successive nights on 14-15 and 15 – 16 April 2000; locust targets were present throughout the radar detection range for approximately 11 hours without any obvious layering. The estimated migration distances were not large. The Thargomindah IMR was probably at the edge of the migration as the locust population was located mainly to the south of Thargomindah, in the Far Northwest of

Figure 5.17 Time-series of Vertical Profile (a) and Displacement Direction (b) for the Night of 15-16 Apr 2000 at Bourke Strong overflight above 500 m agl is evident and lasts the whole night. 2 Estimation of Plague Locust Migrations 171

NSW and the Far North of SA. At Bourke, on the other hand, the IMR detected a very strong overflight with little variation in the displacement direction that lasted the whole night of 15 – 16 April 2000 (Figure 5.17). The locust targets were mainly in the layer between 500 – 1000 m. This migration occurred when Bourke was on the south western side of a decaying high and a cold front was approaching from the southwest (Figure 5.18). Cobar 21-h radiosonde data indicated that a temperature inversion developed at the lowest radar gate level where the temperature was 21.0 °C in comparison to the surface temperature of 17.9 °C there. The temperature was about 16.5 °C at the observed flight ceiling at about 1000 m and 14.5 °C at the top of the radar detection

Figure 5.18 MSL Charts at 10pm and 04 am on the Night of 15-16 Apr 2000 Both Bourke and Thargomindah were affected by the pre-front flow.

Figure 5.19 Trajectories of Chortoicetes terminifera for the Night of 15-16 Apr 2000 Migrating locusts were detected through the whole of night at both Bourke and Thargomindah. Trajectories were for 350-1250 m agl.

172 Chapter 5: Tracing Migration Courses of Australian Plague Locust range. The source region for the locusts was apparently around Bulloo Downs, 300 – 370 km northwest of Bourke, and the destination around Coonamble and Coonabarabran to the southeast (Figure 5.19). This pre-cold-front migration can be expected to have occurred over a large area, transporting large numbers of locusts into the agricultural zone. These populations moved southeastwards into central eastern SA and southern NSW, where large swarm invasions were reported by the APLC. The APLC subsequently sprayed 80 swarms comprising an area of 652 km2 by the APLC during 16 – 28 April 2000, in the Tibooburra – Broken Hill area of NSW and Marree – Yunta area of SA.

The trajectory analysis from IMR observations demonstrates that the technique can be of significant value for locust management by indicating the occurrence of significant migrations and the possible regions into which such populations have emigrated or immigrated. This information can prove invaluable for subsequent locust surveys and early control by the APLC. With the trajectories estimated from the locust migrations detected by two IMRs, the redistributions of locust populations have been clearly illustrated. Following redistribution into favourable habitats, the emigrants were able to successfully breed and a major locust population may develop. Information on the movements and redistribution of locust populations can thus be used operationally to direct field surveys for locust monitoring and for early detection and management of populations that pose a risk to agriculture in eastern Australia.

2.2.2 Seasonal patterns of locust movements detected in 1998 – 2001 With the trajectory analysis, the general trend of plague locust redistributions can be seen from the IMR observations. Due to this species’ capacity for quiescence and diapause and frequent pouplation mixing due to migration, asynchronous generations occur over a wide geographic range and generations boundaries are not distinct. There were no obvious migration waves from different locust generations. Migration trajectories estimated from the identified locust migrations are thus summarised for each season as follows.

In 1998 – 1999, migrating locusts detected at Bourke were generally heading in directions ranging from the west and southwest to the northwest to north, i.e. from the agricultural regions into the arid interior (Figure 5.20). Locust migrations were detected on 52 separate nights between 31 December 1998 and 27 March 1999. No obvious peak 2 Estimation of Plague Locust Migrations 173 period was evident although target numbers were slightly higher after mid-February. Generations cannot be identified from the time series of detected targets.

Forward Trajectory -27 Backward Trajectory

-28

-29

-30

-31 Latitude (deg)

-32

-33

141 142 143 144 145 146 147 148 149 150 151 Longitude (deg) Figure 5.20 Migration Trajectories of Chortoicetes terminifera at Bourke during 1998-1999 Trajectories are calculated for altitudes of 500-650 m agl. Most trajectories are of 7 hr duration.

-27 Forward Trajectory Backward Trajectory

-28

-29

-30

-31 Latitude (deg)

-32

-33

141 142 143 144 145 146 147 148 149 150 151 Longitude (deg) Figure 5.21 Migration Trajectories of Chortoicetes terminifera at Bourke during 1999-2000 Calculation as in Figure 5.20.

174 Chapter 5: Tracing Migration Courses of Australian Plague Locust

In 1999 – 2000, the directions of the locust migrations detected at Bourke were more concentrated than in the previous season, being predominantly toward the west (Figure 5.21). There were 29 nights with detectable locust migrations between 05 November 1999 and 15 April 2000, with intense movements from mid-February to the beginning of March.

In 2000 – 2001, a similar migration pattern of C. terminifera movements to those of 1998-1999 was detected at Bourke (Figure 5.22). There were 54 nights at Bourke with detectable numbers of locusts between 08 November 2000 and 25 March 2001, with population levels generally high after the plague of 1999 – 2000. Many migrations were detected in December, indicating a very high population density that probably arose from overwintering eggs in the southeastern agricultural region. Heavy migrations were also detected in late January to early February 2001, indicating the summer generation was also at very high level. Similar movements occurred in the next generation as well, between the end of February and early March. These continuous movements from the agricultural regions into the arid interior could have contributed to the collapse of the plague in southern NSW and southern areas of SA as there was limited rain in the interior in 2001.

-25

-26 Forward Trajectory -27 Backward Trajectory

-28

-29

-30

-31

Latitude (deg) Latitude -32

-33

-34

-35 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 Longitude (deg) Figure 5.22 Migration Trajectories of Chortoicetes terminifera at Bourke during 2000-2001 Calculation as in Figure 5.20. 2 Estimation of Plague Locust Migrations 175

The migrations of C. terminifera detected by the IMR at Thargomindah, on the other hand, were more diverse, though there were very few towards the southeast (Figure 5.23). Longer flight duration is another characteristic of locust migration observed by the Thargomindah IMR, with migrants often seen through the whole night. The movements in 1999 – 2000 illustrated that populations were accumulating into the arid interior. There were some northward movements, which suggests that plague locusts may have migrated from western NSW to southwestern Queensland.

Forward Trajectory Backward Trajectory -25

-26

-27

-28

-29 Latitude (deg) Latitude

-30

-31

139 140 141 142 143 144 145 146 147 148 Longitude (deg) Figure 5.23 Migration Trajectories of Chortoicetes terminifera at Thargomindah during 1999- 2000 The trajectories are calculated for altitudes of 475-625 m agl. Most trajectories are of 10 h duration as detected often by the IMR at Thargomindah.

The directions of the C. terminifera migrations detected at Bourke were generally concentrated toward the northwest to westward during the study period. This could be due to the prevailing easterly and northeasterly winds in spring and summer (see Chapter 6) and might also reflect a preference by the locusts for particular direction when initiating migration. This indicates that the major migration stream was from the agricultural region to the arid interior in spring and summer. The movements detected at Thargomindah, however, were more diverse, though migrations to the southwest were slightly more frequent than other directions.

176 Chapter 5: Tracing Migration Courses of Australian Plague Locust

Nightly migrations at Bourke were often of 6 – 8 hr duration, which would correspond to typical single night migration distance in the range 200 – 330 km. Figure 5.24 shows the average and standard deviation for the nightly relocation distance at each radar range height. The longest distance migrations occurred where there was usually a nocturnal temperature inversion; the profile of trajectory against flight height illustrates this vertical characteristic.

1250-1400 Forward Backward 1100-1250

950-1100

800-950

650-800

Altitude (m agl) Altitude 500-650

350-500

200-350

100 150 200 250 300 350 400 450 Trajectory Mean Distance (km) Figure 5.24 Trajectory Lengths of Chortoicetes terminifera Detected at Bourke during June 1998 to May 2001 Means and standard deviations for forward and backward trajectories

With IMR observations, the movements of plague locust populations can be monitored and origin and destination regions can thus be estimated. The population variation largely depends on the habitat condition which is strongly associated to rainfall. Whether the metapopulation increases, however, may rely on the success of migrations which commonly occur at the adult stage.

3 Discussion

Although migrations of C. terminifera may be inferred from field-surveys and light-trap catches, migrating locusts can now be directly identified from IMR observations by means of their echo characters with high confidence. The more than 30-40 nights per season of nocturnal migration indicated by the IMR observations seems more frequent 3 Discussion 177 than previously reported for this species (Wright 1987). One of the reasons might be that many migrations that fail to reach suitable habitats go unreported because they don not lead to population build-up. A population exchange between two locations, for example, may not be noticed at all. IMR observations allow the occurrence of a locust migration to be detected and later confirmed with other means, and provide additional information and precision about migration frequency, direction and distance. Therefore, better understanding of its migration mechanisms could be achieved.

Monitoring migration helps understand the mechanisms of population development and distribution of migratory insects. In all three seasons of 1999 – 2001, the IMR at Bourke showed that migrations of plague locust were predominantly from the southeast (see Appendix C). This is presumably a result of locust distributions and prevailing winds. The plague locust habitat map (Figure 5.9) shows that habitats within ‘a single night’s migration range directly to the south or the north of Bourke are predominantly unfavourable for locusts. The main nearby areas of good habitat are to the east and the further away to the southeast in the agricultural region and to the west and southwest in the interior. From the IMR data plague locusts migrated only during periods of disturbed weather, typically associated with of the presence or passage of inland tropical troughs between high pressure systems. Changes in weather patterns with the season would cause westward locust migrations to have a better chance of being detected at Bourke. This would take locusts from the Central Western and North West Plains of NSW into southwestern Queensland. Both regions are very important breeding areas for the species (Key 1945; Wright 1987). Such movements take spring populations from the mainly winter rainfall agricultural zone to the predominantly summer rainfall areas of Queensland; the rapid build up in the interior population from migrants during the season of 1999 – 2000 is such an example. Thargomindah, which is in the interior, could be expected to detect more diverse locust migrations as it is on the boundary between tropical and temperate weather systems (Tapper & Hurry 1993). Migration trajectories estimated from the IMR observations at Thargomindah illustrated this pattern (Figure 5.23). Movements from the south could take locusts from western NSW to western Queensland and locusts from southern Queensland to central western Queensland. Movements to the south also occurred and could be particularly important in autumn for redistributing locusts from the interior to the southern agricultural zone where the probability of winter rain and successful breeding is higher. These bi-

178 Chapter 5: Tracing Migration Courses of Australian Plague Locust directional migrations have had important impacts on the population distributions in all three seasons studied. These northward and southward movements strongly suggest that the locusts have some type of migratory circuit. In addition to the southward autumn migrations emphasised by Wright (1987), the Bourke radar in particular indicates that northward migrations as reported by Reid (1979) and Drake and Farrow (1983) are very common (Deveson et al. 2005) and successful northward migrations in late spring and early summer are an important source in the rapid build up of population in the interior. Such populations can cause severe damage to agriculture in southern and eastern Australia after migrating out from the interior. The sequence of locust migrations and population redistributions in eastern Australia over this period suggests that a locust plague could have started from anywhere within the region, as it is evident the locusts can migrate to avoid temporarily unfavourable environments and dynamically follow available habitats in space and time.

Migration plays an important role in the maintenance and development of populations of C. terminifera in eastern Australia (Wright 1987). Rain is more common in the north of the species’ range during summer and in the south from autumn through to spring. Habitats favourable for C. terminifera survival and breeding are thus available in different regions at different seasons. Northward migrations to habitats in the interior occur in late spring and early summer can be followed by 1 – 2 generations of breeding if sufficient summer rainfall occurs. Southward migrations from the interior during autumn take locusts to the agricultural zone where winter rains generally ensure successful autumn and spring breeding. However, only successful migrants, i.e. those that reach an area with sufficient rainfall and green vegetation, or where rainfall is likely to occur shortly after immigration, will contribute to population development. With successful migrations, the metapopulation of plague locust in eastern Australia can be expected to continuously increase, for example, the locust plague in 2000 was primarily due to the population developments from the northward migrations in early summer and the southward migration in early autumn. Thus potential breeding conditions in destination areas and the ability of locusts to reach such areas through long distance nocturnal migration are key factors in population forecasting.

Using IMRs for monitoring locust migration will significantly improve the efficiency of field-survey and the accuracy of forecasting population development and 3 Discussion 179 distribution from the information on migration routines, origin and destination areas. Field surveys and reports provide the most satisfactory means of confirming migration. A sudden decrease of adult population in an area without evidence of high mortality is most probably due to emigration. A sudden increase of adults in an area without a previously existing population indicates that a major immigration has occurred. Therefore, any migrations detected by the IMRs, and their destination regions predicted by trajectory analyses, can be verified. This will help to prevent outbreak populations developing unnoticed in an unsurveyed area. If a third IMR unit to be added to the network, a location at Broken Hill or in the Northern Agricultural of SA might be preferable. Relocation of the Thargomindah IMR to Windorah, which is at the centre of locust habitats in the arid interior, may also be desirable. Alternatively, relocating the Bourke IMR to Broken Hill, where major migrations between the southeastern agricultural zone and the arid interior can be expected to be detected, would enhance the current network’s coverage.

In addition, IMRs provide a powerful research tool for better understanding migration characteristics such as take-off and flight direction and duration, and the relation of these to weather conditions. From the IMR observations, it is clear that the dusk take-off of plague locust is associated with the change of light intensity but regulated by temperature and wind. The duration of the locust migrations detected at Bourke was normally shorter than that at Thargomindah. Migrations often ceased at Bourke shortly after midnight, but usually lasted the whole night at Thargomindah. Apart from the distance to suitable habitats, warmer temperatures and the presence of nocturnal temperature inversions may be key factors of affecting the duration of locust migration. The observations from an automatic IMR are valuable sources for calculating the trajectories of insect migration, as it does not require additional measurements of upper-wind or fully understanding of the migrants’ orientation behaviour and airspeed. However, with IMR observations, a better understanding can be achieved on the insect flight behaviour (see Chapter 6), and upper winds could therefore be a good estimator of insect migration in application of IMR findings into forecasting, especially when the IMR is off the migration path. The IMR observations confirmed that the plague locusts migrate only under disturbed weather (Clark 1969; Farrow 1979). With this relationship, possible locust migrations can be forecasted based on weather observations. Furthermore, the computer driven IMR operations have accumulated a

180 Chapter 5: Tracing Migration Courses of Australian Plague Locust large amount of data on the time-series of airborne insect fauna, which is also valuable information for systematic study of insect diversity.

The IMRs in eastern Australia have proven valuable in directly monitoring migrations of C. terminifera between the southeastern agricultural region and the arid interior, although their locations prevented them from detecting most movements within the interior and southwest from the interior. The movements between generations are usually conducted by adults in two to five nights of continuous migration, which can cover more than 1000 km (Farrow 1979). The IMR observations at Bourke in particular demonstrated this migration wave pattern. Monitoring and forecasting the movements of locust populations between the agricultural zone and the interior are therefore practicable with the automated IMRs for effective and efficient pest management (Hunter & Deveson 2002).

6 Orientation of Australian Plague Locust

1 MATERIALS AND METHODS ...... 182 1.1 MIGRATION EVENTS...... 182 1.2 ENVIRONMENTAL DATA...... 182 1.3 STATISTICAL ANALYSIS...... 186 2 EXAMINATION OF ORIENTATION MECHANISMS...... 186 2.1 EVIDENCE FOR ORIENTATION IN NOCTURNAL FLIGHT...... 187 2.2 TRUE NAVIGATION ...... 189 2.2.1 Topographic Features...... 189 2.2.2 Environmental Gradients...... 191 2.3 VECTOR NAVIGATION...... 191 2.3.1 The Earth’s Magnetic Field ...... 192 2.3.2 The Moon Azimuth ...... 192 2.4 ORIENTATION TO THE WIND ...... 195 2.4.1 Downwind Flight...... 195 2.4.2 Upwind Flight ...... 198 2.4.3 Orientation to the Wind...... 200 3 DISCUSSION...... 205

IMR is not only a useful tool in the practice of insect-migration monitoring (see Chapter 5), but also a powerful device in the research of insect-flight behaviour (Riley 1979). Australian plague locusts Chortoicetes terminifera have often been observed with mutual orientation in nighttime flights (Drake 1983). However, little is known about the orientation mechanisms (Riley & Reynolds 1986; Riley et al. 1988; Riley 1989, 1999), though collective orientation is very common among nocturnal flying insects (Schaefer 1970; Roffey 1972; Riley 1975; Schaefer 1976; Drake 1983; Riley & Reynolds 1986). Its importance is also little understood, but it will have some effect on migration paths and thus the persistence of migratory populations in heterogenous habitats. Therefore, a better understanding would improve the accuracy with which migrant sources and destinations can be located, as well as the understanding about insect migration.

181 182 Chapter 6: Orientation of Australian Plague Locust

To explore the orientation mechanisms of Chortoicetes terminifera, the observed orientation phenomena with IMR are to be related to available external cues, such as the moon’s azimuth and the earth’s magnetic field, and to other environmental factors, such as the wind. Circular statistics are used to examine the orientation behaviours and their relationships to the environmental cues.

1 Materials and Methods

1.1 Migration events

101 nights have been selected from the 140 observed migration nights of Chortoicetes terminifera at Bourke between 1 June 1998 and 31 May 2001 (see Appendix B), based on the selection criteria:

the good-quality echoes (see Chapter 3) of the Australian plague locust within the detection range were more than 15 during 20 – 21 h EST (Eastern Standard Time, for all time quoted hereafter);

both displacement directions and orientation alignments of plague locust were concentrated at preferred angles, unimodally and bimodally distributed respectively with P-value < 0.05 of the Rayleigh test (see Appendix D.3).

1.2 Environmental data

The earth’s magnetic field The earth’s magnetic field has the characteristics of direction and gradient features. The magnetic intensity and declination of southeastern Australia are plotted as contours of total intensity (nano-Telsa) and variation (degree) in the Miller projection (Figure 6.1). The magnetic declination (magnetic variation) and other magnetic field elements for any given geographic coordinates can be calculated with the calculator on the Geoscience Australia web site (http://www.ga.gov.au). The magnetic declination at Bourke was about +9.6° in the middle of the study period (i.e. magnetic north is 9.6° east of true north), and the isogonic lines lie in the direction of northeast-southwest. The 1 Materials and Methods 183 strength of magnetic field ranges from 50 to 60 micro-Teslas, and the isomagnetic contours lie along a northwest-southeast direction.

Figure 6.1 Isomagnetic (a) and Isogonic (b) Maps of Southeastern Australia The total intensity and the magnetic declination of magnetic field is produced by the IGRF-applet of Korhonen (2000), based on the International Geomagnetic Reference Field 2000 model.

The moon’s azimuths The moon is the obvious reference at night. The moon’s azimuths and elevations for the nights with detected locust migration were calculated for each hour for the location of Bourke with the online calculator at the Astronomical Applications Department of the U.S. Naval Observatory (http://www.aa.usno.navy.mil/). Fraction of the moon illuminated was also calculated. Whether the moon was visible when it was above the horizon, however, was not able to verify due to the absence of cloud record. Nevertheless, the moon was assumed to be visible if it was above the horizon, as rain is not often at Bourke, unless rain was detected by the rain detector incorporated with the IMR unit.

Radiosonde upper winds The radiosonde data for 21 h, at Cobar in New South Wales was provided by the Australian Bureau of Meteorology for the period of the IMR observation. Cobar station (31.4853° S, 145.8292° E, 260 m MSL) is 160 km to the south of Bourke.

184 Chapter 6: Orientation of Australian Plague Locust

Simulated upper winds Simulated upper winds can be interpolated for any geographic location at any time from the 6-hourly analysis outputs of the Limited Area Prediction System (LAPS), a regional atmospheric circulation model developed and used by the Australian Bureau of Meteorology (Puri et al. 1998). A comparison with the Cobar radiosondes was done for validation purpose. Simulated winds were calculated for Cobar for 21 h at eight heights of 150-m interval, starting at 275 m geopotential height. The radiosonde upper winds were linearly interpolated from the measures to the same heights as those of the simulated winds. The differences between simulated winds and radiosonde measures were tested by two-sample paired t-test on both speed and direction after the subtraction that converts the directional data into linear scale (Batschelet 1981). The differences were calculated as the subtraction of the simulated values by the radiosonde values (Table 6-1). The results show that the simulated winds are generally close in direction but slightly slower in speed to the radiosonde data at Cobar (Figure 6.2). It is obvious that simulated winds are similar to radiosonde data at all heights except for the lowest one, which is about 15 m above ground level and at which the simulated winds are much faster than the radiosondes.

Table 6-1 Difference of Simulated and Radiosonde Winds at Cobar

GeoHeight (m) 275 425 575 725 875 1025 1175 1325

Mean 3.39 -0.60 -1.52 -1.34 -0.99 -0.78 -0.69 -0.62 Speed SD 2.23 2.53 2.60 2.29 2.07 1.86 1.73 1.75 (m/s) P-value 0 0.0001 0 0 0 0 0 0

Mean -10.3 -1.8 0.5 -0.2 1.9 0.1 -0.2 -1.1 Direction SD 39.6 36.4 36.2 34.8 35.7 33.6 32.0 28.2 (deg) P-value 0.0268 0.0022 0.0058 0.7228 0.8696 0.9895 0.6269 0.1679

Night 290 290 290 289 289 289 288 287 Differences of wind speed and wind direction are at eight geopotential heights between 275 and 1325 m. A negative value of direction difference means the simulated direction is left-shifted (counterclockwise) from the radiosonde. Data are for the period of 1 Jan – 30 Nov 2000.

Although the simulated wind speeds are significantly different from those actually measured at Cobar, the difference maxima are mostly less than 1.5 m/s. The simulated wind directions, on the other hand, are within 10° of the measured values, and 1 Materials and Methods 185 are not significantly different at high levels. The differences can be attributed to the limitation of the LAPS models’ ability to simulate wind fields at a small scale from large-scale data. The lack of simulation of the nocturnal temperature inversion at lower levels might also be a source of error, which is evident from the profile of speed difference. Nevertheless, simulated winds are good approximations to radiosonde values.

60 1325m Total: 287 Total: 287 30 0 60 1175m Total: 288 Total: 288 30 0 60 1025m Total: 289 Total: 289 30 0

60 875m Total: 289 Total: 289 30 0

60 725m Total: 289 Total: 289 30 0

60 575m Total: 289 Total: 290 Number of Nights Number of 30 0

60 425m Total: 290 Total: 290 30 0

60 275m Total: 289 Total: 290 30 0 -8-4048-180 -120 -60 0 60 120 180 Difference of Wind Speed (m/s) Difference of Wind Direction (deg) Figure 6.2 Comparison of Simulated and Radiosonde Winds at Cobar, NSW Dataset is same as in Table 6-1.

The simulated winds for Bourke were calculated for 21 h at the eight heights from 275 m above ground level with 150 m interval, i.e. from 382 to 1432 m of geopotential height, corresponding to the med-points of the IMR sampling ranges. The difference between the geopotential and geometric heights can be ignored as at maximum value over the Bourke IMR operating range is 2.1 m.

Ground weather observations The surface winds and temperatures were recorded at the intervals of 5 or 6 min by the automatic weather station incorporated with the IMR at Bourke.

186 Chapter 6: Orientation of Australian Plague Locust

1.3 Statistical Analysis

The distributions of the displacement direction and the orientation alignment for the Australian plague locusts detected in the 2-h period of 20-21 h were tested using the Rayleigh test for uniformity (Batschelet 1981). The Rayleigh test is parametric for the significance of the mean direction and the concentration parameter of a circular distribution. The mean angle and the standard deviation were calculated if the distribution is unimodal for displacement, or after doubling the unimodal angles for orientation (for which is axial) at significance level P < 0.05. The median angles were also computed. However, they were often identical to the mean angles as the angle distributions are concentrated in narrow ranges. Therefore, mean values were used in all analyses.

The homogeneity of two samples was tested using Kuiper-Stephens K* test and Watson-Stephens U2 test (Upton & Fingleton 1985). Both tests are non-parametric. They can be used for the goodness-of-fit test for a von Mises distribution, the circular analog of the normal distribution on a line, or any particular distribution.

For statistics, the locust data were summarised for eight altitude ranges between 200 – 1400 m above ground level at 150-m intervals, using the 2-hr IMR observations of 20 – 21 h during the 3 years at Bourke. The Rayleigh uniformity test was applied to each of the eight subset data and only the direction concentration of both the orientation and the displacement directions at P < 0.05 level were included for counts.

Further details and computational implementation are given in Appendix D.3.

2 Examination of Orientation Mechanisms

It seems unlikely that nocturnal flying insects could use visual or audio references to each other to orientate collectively, since they are usually far apart in the night sky (Riley & Reynolds 1986). The frequent observations of collective orientation therefore suggest that the Australian plague locusts adjust their headings relatively to one or more external cues. The IMR observations of plague locust orientation will be related to environmental factors in an attempt to identify the mechanism of orientation. 2 Examination of Orientation Mechanisms 187

2.1 Evidence for Orientation in Nocturnal Flight

Radar studies of night-flying insects have often revealed uniform headings (Riley 1973, 1975; Drake 1983), indicating that many insects have an ability to orientate themselves to some cue. The IMR observations indicated that Australian plague locusts frequently exhibited mutual orientation. Figure 6.3 shows that similar displacement directions and body alignments were observed at all radar range gates during an intense migration of Australian plague locusts over the Bourke IMR through the whole night of 08-09 March 2001. From the total of 4111 good-quality echoes, the overall displacement direction was 316.3±17.3° with mean resultant length (the vector mean) R = 0.90 . The average ground speed on this occasion was 15.3±3.1 m/s, and thus the deduced heading (assuming downwind migration) would be 351.0±22.3° ( R = 0.71) as at this displacement speed they would have flown downwind. The crab angle was 33.3±46.9°. These locusts were travelling northwestward while heading roughly northward; the alignment distributions differed only slightly over the 200-1400 m altitude range.

(a) N (b) N (c) N (d) N 200-350m 350-500m 500-650m 650-800m Total: 807 Total: 970 Total: 789 Total: 582

(e) (f) (g) (h) 800-950m 950-1100m 1100-1250m 1250-1400m Total: 636 Total: 191 Total: 105 Total: 31

S S S S Figure 6.3 Orientation and Displacement Direction Profiles of Chortoicetes terminifera at Bourke on 08-09 Mar 01 Equal-area polar histograms of orientation (pink, axial) and displacement (blue, unimodal) directions for plague locusts at each of 8 height ranges between 200 and 1400 m (a – h). Counts are for 21 to 02 h.

Uniform alignment over a wide altitude range and lasting several hours, as in Figure 6.3, however, had not been observed very often. Instead, the locusts were more often seen heading in different directions at different times and different altitudes. In

188 Chapter 6: Orientation of Australian Plague Locust fact, locusts always exhibited mutual orientation over short time periods and shallow altitude ranges. Figure 6.4 shows an example of locusts shifting their headings counterclockwise at high altitudes; headings were northward with large crab angles at low level but almost westward with very small crab angles at high levels.

(a) N (b) N (c) N 200-400m Total: 826500-700m Total: 692 800-1150m Total: 259

Figure 6.4 Orientation and Displacement Direction Profiles of Chortoicetes terminifera at Bourke on 28 Feb 1999 Equal-area polar histograms of orientation and displacement directions at three height ranges, for the period 19 – 22 h.

N 1400

1200

1000 10 m/s

800

600 Altitude (m)

400

200 Bourke, 25-26 Feb 2001 Displacement Orientation (7673 targets) 0 19h 20h 21h 22h 23h 00h 01h 02h 03h 04h 05h Time (h) Figure 6.5 Orientations and Displacement Directions of Chortoicetes terminifera at Bourke on 25-26 Feb 2001 The mean displacements (black arrows) and body alignments (red bars) are plotted from averages for each 1-hr interval of observation, for samples when more than 15 plague locust targets were detected and orientations or displacements showed non-uniformity at P < 0.10 level by the Rayleigh test. 2 Examination of Orientation Mechanisms 189

The IMR observations showed that plague locusts always fly with some degree of collective orientation. Figure 6.5 shows an example of locust orientations changing with both altitude and time. The overall directions of both displacement and orientation spread over a wide angle. But within each small (150 m) altitude range and short (1 hr) time period, the locusts had both their body alignments and their displacement directions concentrated into a narrow range of angles. This illustrated that the locusts always orient themselves during migration, suggesting that the locusts react in similar ways to certain environmental cues.

2.2 True Navigation

For insects, there is no species that completes true return migration, i.e. the same individuals make round-trip journeys. Therefore, it is unlikely that they develop navigation mechanisms based on experience (Bingman & Cheng 2005). However, during the long history of evolution and adaptation under the nature selection, whether the Australian plague locusts migrate in response to definite landscape features or variations in the distribution of chemical cues or in the earth’s magnetic field, will be examined here.

2.2.1 Topographic Features The most important terrestrial feature near the Bourke radar site is the Darling River, a major regional drainage, which flows southwestward in this region. The level rises about 150 m to the southeast (the Cobar Peneplain) and to the southwest (the Broken Hill Upland), but not sufficiently to form a wind tunnel along the Darling River (see Figure 2.1 in Chapter 2). The locusts did not align themselves along the NE-SW direction of the river (Figure 6.6). Instead, there were more locust movements with northwestward headings than that in any other directions. Thus the major river system in southeastern Australia seems to have no impact on the locusts’ headings.

The displacement directions also show no evidence of the locusts following the NE-SW alignment of the river (Figure 6.7). At low altitudes the locusts flew predominately northwestward, which may be attributed to the prevailing winds during the locust migration season (see Figure 6.8). Generally, southeastly winds were more frequent than the opposite direction at low levels, but more easterly winds at high levels.

190 Chapter 6: Orientation of Australian Plague Locust

(a) 0 200-350m (b) 0 350-500m (c) 0 500-650m (d) 0 650-800m Total: 78 Total: 85 Total: 83 Total: 88

270 90 270 90 270 90 270 90

180 180 180 180

(e) 0 800-950m (f) 0 950-1100m (g) 0 1100-1250m (h) 0 1250-1400m Total: 75 Total: 74 Total: 70 Total: 44

270 90 270 90 270 90 270 90

180 180 180 180 Figure 6.6 Orientations of Chortoicetes terminifera Detected at Bourke during 1998-2001 Nights are counted when both the orientation and displacement directions were significantly concentrated (P < 0.05) at preferences during the 2-hr observations of 20-21 h at each 150-m interval from locust migration events detected at Bourke during the three seasons.

(a) 0 200-350m (b) 0 350-500m (c) 0 500-650m (d) 0 650-800m Total: 78 Total:85 Total: 83 Total: 88

270 90 270 90 270 90 270 90

180 180 180 180

(e) 0 800-950m (f) 0 950-1100m (g) 0 1100-1250m (h) 0 1250-1400m Total: 75 Total: 74 Total: 70 Total: 44

270 90 270 90 270 90 270 90

180 180 180 180 Figure 6.7 Displacement Directions of Chortoicetes terminifera Detected at Bourke during 1998-2001 Dataset is same as in Figure 6.6.

This is because that Bourke was often under the influence of tropical troughs within which the plague locusts conduct their nocturnal migrations (see Section 1.3 in Chapter 5). In addition, the optomotor response (Riley et al. 1988) may not play an important role in locust orientation at nighttime, as no obvious differences of vegetation are around the Bourke region and high flying locusts may not be able to detect the change of ground pattern. 2 Examination of Orientation Mechanisms 191

0 0 0 0 (a) Total: 78 (b) Total: 85 (c) Total: 83 (d) Total: 88 275m 425m 575m 725m

270 90 270 90 270 90 270 90

180 180 180 180

(e) 0 (f) 0 (g) 0 (h) 0 875m Total: 75 1025m Total: 74 1175m Total: 70 1325m Total: 44

270 90 270 90 270 90 270 90

180 180 180 180 Figure 6.8 Distributions of Wind Bearing during Locust Migrations at Bourke during 1998- 2001 Simulated wind bearings corresponding to the dataset as in Figure 6.6

Therefore, the locusts may not be able to sense the landmark features at high altitude at night and orientate to them.

2.2.2 Environmental Gradients Although the dominant direction (in SE – NW, Figure 6.6) of locust body alignments lies more or less parallel to the isomagnetic lines (Figure 6.1a) of the earth’s magnetic field, the wide variations in the distribution of detected locust headings with time and altitude suggest that the locust do not align their bodies along the gradient of earth’s magnetic field, or at any fixed angle to this direction.

There are not obvious variations of any chemical gradient in this region. Therefore, the plague locust might not have the ability of true navigation.

2.3 Vector Navigation

Whether a migrant insect can maintain a pre-determined orientation to reach a migratory destination for a specific period of time or distance is called vector navigation (Bingman & Cheng 2005). A variety of celestial cues (sun, stars, moon, and sky polarisation) and the earth’s magnetic field could be used as potential directional references for the whole or any legs of vector navigation. The variation of the angle between the locusts’ headings and the displacement directions is checked against some compass cues.

192 Chapter 6: Orientation of Australian Plague Locust

2.3.1 The Earth’s Magnetic Field The earth’s magnetic field is potentially a reliable direction cue. The annual movement of the earth’s magnetic poles is very short, currently about 40 km a year (Natural Resources Canada 2008), thus the magnetic variation at any geographic location changes very little. Sea animals and birds, for example turtles and pigeons, and daytime flying insect monarch butterflies, have been found to have the capacity to use the earth’s magnetic compass for navigation (Perez et al. 1999; Walker et al. 2002). Whether the locusts have such ability, and use it during nocturnal migration is not evident from the IMR observations.

If the locusts could sense the earth’s magnetic field, they would use either the direction (polarity and inclination) or the intensity as navigational cues. Neither the locust orientations (Figure 6.6) nor the locust displacements (Figure 6.7) were aligned with either of these directions (Figure 6.1).

2.3.2 The Moon Azimuth At night time, the moon is an obvious directional reference. Whether the plague locusts use the moon as an orienting compass, is examined with the IMR observations.

It is not likely that the orientation of plague locust is affected by the moon’s changing azimuth during nocturnal flights (Figure 6.9), on a night then the moon age

N 1400 Bourke, 08-09 Mar 2001 (60-760 mg, 8382 targets) 1200

10 m/s 1000

800

600 Altitude (m)

400

200 Displacement Orientation Moonlight 0 19h 20h 21h 22h 23h 00h 01h 02h 03h 04h 05h Time (h) Figure 6.9 Orientations of Chortoicetes terminifera Observed at Bourke on 08-09 Mar 2001 The blue arrows show the moon azimuthal directions. 2 Examination of Orientation Mechanisms 193 was 13 days (one day before full moon) and the fraction illuminated was 98%. As the moon went up and down from northeast to northwest during the night, the locust orientations and displacement directions changed little, flight was to the northwest with almost northward headings throughout this period.

There was no clear relation between the locust headings and the azimuths of moonlight direction. Figure 6.10 shows the distributions of the acute angles between the moon’s azimuth and the locust’s body alignment. It has no obvious clusters. Randomness tests indicate that uniformity cannot be rejected even at significance level P = 0.2 (Table 6-2). In addition, the angle distributions of the locust orientation to the moon’s azimuth at different radar gate ranges were not significantly different (P > 0.15 for both the Kuiper’s and the Watson’s two-sample tests of homogeneity).

1250-1400m 2

0 4 1100-1250m 2 0 4 950-1100m 2 0 4 800-950m 2 0 8 650-800m 6 4 2 0 NumberNights of 4 500-650m 2 0 4 350-500m 2 0 4 200-350m 2 0 -90 -60 -30 0 30 60 90 Angle of Orientation to the Moon's Azimuth (deg) Figure 6.10 Angles of Body Alignment of Chortoicetes terminifera to the Moon’s Azimuth The angles between the orientation direction and the moon’s azimuth are for the nights when the moon was above the horizon. A negative angle means moonlight is from the right side of the locust’s heading.

Table 6-2 P-Values of Randomness Test for Angles of Locust Orientation to the Moon Azimuth Range Gate A B C D E F G H Rayleigh 0.675 0.341 0.247 0.381 0.370 0.949 0.770 0.718 Watson’s 0.535 0.261 0.219 0.322 0.534 1.000 0.958 0.888 Nights 41 45 45 45 37 38 34 21

194 Chapter 6: Orientation of Australian Plague Locust

The relationship between the locust displacement directions and the moon’s azimuth, on the other hand, appears non random (Figure 6.11, P = 0.1, Table 6-3). There were more nights when the moon was on the left side of the locust displacement directions. However, this could be a result of the non-random distributions of the locust displacement directions detected at Bourke (Figure 6.7) and the nature of seasonal migrations in association with the moon’s periodic cycle.

2 1250-1400m

0 1100-1250m 2

0 950-1100m 2

0 4 800-950m 2 0 4 650-800m 2 0 Number of Nights Number of 4 500-650m 2 0 4 350-500m 2 0 4 200-350m 2 0 -180 -150 -120 -90 -60 -30 0 30 60 90 120 150 180 Angle of Displacement to the Moon's Azimuth (deg) Figure 6.11 Angles of Displacement of Chortoicetes terminifera to the Moon’s Azimuth Negative angle means the moon is on the right side of the locust. Same data set as in Figure 6.10.

Table 6-3 P-Values of Randomness Test for Angles of Locust Displacement to the Moon Azimuth

Range Gate A B C D E F G H

Rayleigh 0.018 0.076 0.022 0.031 0.033 0.035 0.092 0.776

Watson’s 0.026 0.111 0.029 0.047 0.051 0.051 0.105 0.878

Nights 41 45 45 45 37 38 34 21

2 Examination of Orientation Mechanisms 195

2.4 Orientation to the Wind

It has been widely reported that insects can maintain downwind or crosswind headings during nocturnal flight (Schaefer 1976; Drake 1983; Riley & Reynolds 1983, 1986). A sudden change in wind directions is therefore expected to produce a similar change in orientations. The passage of a cold front during the night of 22-23 Dec 2000 provided an example of this situation (Figure 6.12). The front was identified on Bureau of Meteorology weather charts, and its arrival at the Bourke radar site at 22:40 h timed from sudden changes in the winds and temperatures recorded by the AWS at the IMR site. With the passage of the front, displacement directions changed from northerly or northwesterly to southerly or southeasterly. Orientations changed from eastward to northward.

N 1400

1200

1000 10 m/s

800

600 Altitude (m)

400

200 Bourke, 22-23 Dec 2000 Displacement Orientation (13609 targets) 0 19h 20h 21h 22h 23h 00h 01h 02h 03h 04h 05h Time (h) Figure 6.12 Orientations and Displacement Directions of Chortoicetes terminifera at Bourke on the Night of 22-23 Dec 2000 A cold front passed on Bourke at about 22:40 h.

2.4.1 Downwind Flight IMR observations showed that locust movements at Bourke were largely determined by the wind. Figure 6.13 shows the direction of locust displacement in relation to wind bearing. On most nights the locusts flew approximately, though not exactly, downwind.

196 Chapter 6: Orientation of Australian Plague Locust

360 (a) (b) 270

180

0 (c) (d) 270

180

0 (e) (f) 270

180

Displacement Direction (deg) Direction Displacement 90

0 (g) (h) 270

180

0 0 90 180 270 0 90 180 270 360 Wind Bearing (deg) Figure 6.13 Displacement Directions of Chortoicetes terminifera at Bourke in Relation to Wind Directions Displacement directions are averaged from the 20-21 h IMR observations at altitudes of 200-1400 m agl (a-h, 150-m interval) from the 101 nights of locust migration. Winds were simulated for 21 h for the same heights as the IMR observations. Lines show exactly downwind displacement directions. 2 Examination of Orientation Mechanisms 197

The locust displacements were not symmetrically spread on either side of the wind direction, presumably as result of their orientation behaviour. Displacement directions were more often to the right of the wind than to the left; for example, the locusts travelled northwestward in easterly wind.

Figure 6.14 shows the differences between the locust displacements and the winds in histogram form. The locusts travelled mostly faster than the winds and within 90° of the wind bearings, with a trend towards the right. The small number of upwind movements may be due to inaccurate wind simulation.

10 1250-1400m Total: 44

10 1100-1250m Total: 70 0

10 950-1100m Total: 74

10 800-950m Total: 75

10 650-800m Total: 88

Number of Nights of Number 0

10 500-650m Total: 83 0

10 350-500m Total: 85 0

10 200-350m Total: 78

0 -12 -8 -4 0 4 8 12 -180 -120 -60 0 60 120 180 Speed Difference of Wind to Displacement (m/s) Direction Difference of Wind to Displacement (deg) Figure 6.14 Differences of Displacement of Chortoicetes terminifera and Wind at Bourke The differences of speed and direction between the locust displacements and the simulated winds are calculated by subtracting the displacement from the wind vectorially, for occasions when the locusts showed a concentrated distribution of displacement directions. Dataset is same as in Figure 6.13.

The plague locusts were observed at Bourke with more right-shifted than left- shifted headings. Figure 6.15 shows the histograms of crab angle distribution of detected concentrated locust migration nights. Although the distributions of crab angle were not uniform at all altitudes, they were not all unimodal and thus cannot be fitted with the von Mises distributions at P = 0.05 level. However, the right-shifted headings may be related to the predominance of westward and north-westward winds in the locust dataset.

198 Chapter 6: Orientation of Australian Plague Locust

1250-1400m 10 Total: 44

0 1100-1250m Total: 70 10

0 950-1100m 10 Total: 74

0 800-950m Total: 75 10

0 650-800m 10 Total: 88

0 Number of NightsNumber of 500-650m Total: 83 10

0 350-500m 10 Total: 85

0 200-350m Total: 78 10

0 -90 -60 -30 0 30 60 90 Angle of Orientation to Displacement (deg) Figure 6.15 Distributions of Crab Angles of Chortoicetes terminifera at Bourke 0 ~ –90° is left-shifted heading, 0 ~ 90° is right-shifted heading. Dataset is same as in Figure 6.13.

2.4.2 Upwind Flight Upwind flights were rarely if ever seen. Figure 6.14 suggests that there were a few occasions when locusts might have flown against the wind. In Figure 6.16, the variation of the angle between displacement and simulated wind direction with wind speed is shown. It can be seen that when the wind was faster than 5 m/s, the locusts flew within 90° of the wind. Upwind flights were associated with weak winds. This is reasonable and is supported by many studies (Riley 1975; Riley & Reynolds 1979; Riley et al. 1983). However, these upwind flights may not all be true as some occurred when a weather system were passing over Bourke, and the simulated winds may then be inaccurate. In fact, on those nights when the locusts appeared to be flying upwind, weather systems of this type were present. For example, lows close to Bourke on the nights of 09-10 January 1999, 07-08 January 2001, 13-14 January 2001 and 25-26 February 2001 accounted for 7 nights on Figure 6.16. 2 Examination of Orientation Mechanisms 199

1250-1400m 20 Total: 44 10

0 1100-1250m 20 Total: 70 10

0 950-1100m 20 Total: 74 10

0 800-950m 20 Total: 75 10

0 650-800m 20 Total: 88 10

Wind (m/s) Speed 0 500-650m 20 Total: 83 10

0 350-500m 20 Total: 85 10

0 200-350m 20 Total: 78 10

0 -180 -120 -60 0 60 120 180 Angle of Displacement to Wind (deg) Figure 6.16 Displacement Directions, Relative to the Wind, of Chortoicetes terminifera, in Relation to Wind Speed The simulated wind speeds are plotted against the angles between the locust displacements and the wind bearings.

200 Chapter 6: Orientation of Australian Plague Locust

Assuming Australian plague locusts migrate downwind at night and their airspeeds are proportional to their masses (Dudley 2000a), the upper winds could be estimated vectorially from the IMR displacements and headings and the assumed airspeeds. Figure 6.17 shows the comparison of the estimated winds from the IMR observations with the simulated winds from the LAPS model by calculating the differences of speed and direction. It is found that when the mean airspeed of locust was 6.5 m/s, from vm= 0.36 where m is mass (mg), the estimated winds were very close to those simulated by the model. However, there were a few nights when the estimated winds differed greatly from the simulated ones, which means either the locusts flew upwind then or the simulated winds were in wrong directions. Nonetheless, it is clear that the locusts flew mostly downwind.

10 1250-1400m Total: 44

0 10 1100-1250m Total: 70

0 10 950-1100m Total: 74

0 10 800-950m Total: 75

10 650-800m Total: 88

10 500-650m Total: 83 Number of Nights

10 350-500m Total: 85

10 200-350m Total: 78

0 -12 -8 -4 0 4 8 12 -180 -120 -60 0 60 120 180 Difference of Speed (m/s) Difference of Direction (deg) Figure 6.17 Differences of Simulated and Estimated Wind at Bourke The simulated winds are compared with the estimated winds calculated vectorially from the 2-h IMR observations (as in Figure 6.14) and the theoretical locust airspeed of 6.5 m/s.

2.4.3 Orientation to the Wind In this section the relation of locust orientation to wind direction is investigated further. Figure 6.18 shows the distributions of the angle between the locust orientation and the wind direction. These angles are not uniform at any height (P < 0.001 for both the 2 Examination of Orientation Mechanisms 201

Rayleigh and the Watson uniformity tests). The distributions are unimodal at all heights and similar in form and location, and are fitted well with a von Mises distribution (mean = 60°, resultant length = 0.89, P > 0.15). This means that there were more nights when the locusts flew with a tailwind from the right than from the left or from directly behind. This strongly suggests that the locusts select heading directions that are to the right of the wind. However, the wide range of the distributions of their orientation angles relative to the wind indicates that the wind may not be the only cue affecting locust orientation during nocturnal migration.

1250-1400m 10 Total: 44 0 1100-1250m 10 Total: 70 0 950-1100m 10 Total: 75 0 800-950m 10 Total: 88 0 650-800m 10 Total: 88

Number of Nights Number 0 500-650m 10 Total: 83 0 350-500m 10 Total: 85 0 200-350m 10 Total: 78 0 -90 -60 -30 0 30 60 90 Angle of Orientation to Wind (deg) Figure 6.18 Angle Distributions of Locust Orientation Relative to Wind Bearing The direction was calculated by subtraction of wind direction from the locust heading direction. A positive angle means the locusts flew within the right tailwind (i.e. the locust heading direction is to the right of the direction wind is blowing towards). Dataset is same as in Figure 6.14.

Locust orientation is unrelated to wind speed (Figure 6.19). The orientation angles of the locusts to the wind directions did not decrease as the wind intensity increased, nor did it increases. The similar angles of the orientation to the wind might have occurred under both nearly calm and windy conditions, though the upper winds were often strong during locust migrations.

202 Chapter 6: Orientation of Australian Plague Locust

90 (a) (b) 60 30 0 -30 -60 -90 90 (c) (d) 60 30 0 -30 -60 -90 90 (e) (f) 60 30 0 -30

Angle of Orientation to Wind to Angle of Orientation (deg) -60 -90 90 (g) (h) 60 30 0 -30 -60 -90 0 4 8 1216200 4 8 121620 Wind Speed (m/s) Figure 6.19 Orientations of Chortoicetes terminifera Relative to Wind Speeds A positive angle means a right-shifted heading. Dataset is same as in Figure 6.14.

The locusts seemed to orientate to the wind directions in some degree (Figure 6.20). It is obvious that the angles of orientation to wind were not random from the clusters of scatter plots. Easterly and southeasterly winds (between 180 and 360° wind 2 Examination of Orientation Mechanisms 203

90 60 30 0 (a) (b) -30 -60 -90 90 60 30 0 (c) (d) -30 -60 -90 90 60 30 0 (e) (f) -30 Angle of Orientation to Wind (deg) -60 -90 90 60 30 0 (g) (h) -30 -60 -90 0 0 0 0 0 0 0 8 -9 90 60 -9 90 -1 180 270 3 -18 18 270 360 Direction of Wind Bearing (deg) Figure 6.20 Orientations of Chortoicetes terminifera Relative to the Wind Direction Plotted against Wind Direction A positive angle means right-shifted heading. Points for wind directions (open squares) between 180 and 360° have been duplicated (with open circles) into the range of -180 ~ 0°. Dataset is the same as for Figure 6.14.

204 Chapter 6: Orientation of Australian Plague Locust

90 (a) (b) 60 30 0 -30 -60 -90 90 (c) (d) 60 30 0 -30 -60 -90 90 (e) (f) 60 30 0 -30 -60 -90

Angle of Displacement to Orientation (deg) Orientation of Displacement to Angle 90 (g) (h) 60 30 0 -30 -60 -90 0 0 0 -90 90 80 -90 90 -180 1 270 -18 180 270 360 Displacement Direction (deg) Figure 6.21 Crab Angles of Chortoicetes terminifera Plotted against Displacement Directions Dataset is same as in Figure 6.20. bearings) predominated, and the locusts had predominately right-shifted headings (above the horizontal line of 0 in the vertical axis) in these conditions. However, there 2 Examination of Orientation Mechanisms 205 were no clear correlations between the locust orientations and the wind directions. The crab angles of orientation relative to displacement (Figure 6.21), on the other hand, had a much clearer pattern with the displacement directions. The locusts may have adjusted their headings to compensate their body alignments to the wind. Nonetheless, further observations must be done before a conclusion can be drawn.

Figure 6.22 shows the crab angles related to the displacement directions. As the direction of displacement changed, the crab angle also changed. This may imply that the locusts orientate not only to the wind, but also to some other cues with a fixed directional reference.

200-350m 350-500m 500-650m 650-800m (a) N Total: 78 (b) N Total: 85 (c) N Total: 83 (d) N Total: 88 90

-90

90 800-950m 950-1100m 1100-1250m 1250-1400m (e) N Total: 75 (f) N Total: 74 (g) N Total: 70 (h) N Total: 44 90

-90

90 Figure 6.22 Polar Plot of Crab Angles in Relation to Displacement Directions for Chortoicetes terminifera at Bourke Displacement directions are shown as angles, with N to the north. Crab angles are shown as radii ranging from –90° at the centre to +90° at the periphery. Dataset is same as in Figure 6.14.

3 Discussion

Orientation has been detected often among migrating insects. Australian plague locusts have demonstrated similar responses to the same environmental cues. Downwind migration with a heading of a varied angle to the wind suggests that locusts might have orientated with cues in addition to wind. However, investigation has not found the locust orientation in relation to any directional references. Figure 6.23 shows the overall

206 Chapter 6: Orientation of Australian Plague Locust orientation scheme for Australian plague locusts, based on the IMR observations at Bourke.

Wind Bearing Heading

Figure 6.23 Possible Orientation Scheme of Chortoicetes terminifera Abstracted from Figure 6.22.

It is, however, certain that the nocturnal orientation of flying plague locusts is determined partially by the wind. Due to limited locust migration nights observed by the IMR, specifically fewer southward migrations, orientation to the wind seemed strongly related but unfortunately can not be drawn to a conclusion of relationship (Figure 6.22). The strong patterns of orientation in the displacement direction from southeast to northwest might have resulted from most locust migrations observed at Bourke in westerly directions. This however, apart from the location of the IMR at Bourke related to the locust occurrence area, locusts may have chosen preferred winds under disturbed weather to migrate (see Chapter 5). It is evident that the displacement directions of locust migration were not random but dominantly towards the northwest.

Orientation behaviour usually alters flight trajectory and therefore affects migration path. Monarch butterflies Danaus plexippus migrate along a fixed route and have different orientation strategies in different places, which seems an adaptive behaviour under natural selection pressure (Gibo 1986; Perez et al. 1999). Desert Locusts are believed to have adapted to ride favourable winds into potential rainy areas 3 Discussion 207 in the subtropical deserts of northern Africa and Southwest Asia (Riley & Reynolds 1996). Beet armyworm moths Spodoptera exigua always migrate with headings toward inland avoiding the sea, regardless of the prevailing wind directions (Feng et al. 2003). IMR observations showed that Australian plague locusts migrated predominantly downwind, although their orientation cannot be simply related to either the wind direction or the wind speed. There is no convincing evidence that the locusts orient to any specific geographic reference.

Adaptive behaviour may affect insect orientation. Orientation could be a species-dependent behaviour, possibly an adaptation to environmental conditions that help each species to reach favourable habitats, or it may be a simpler passive response to the flight environment (Zhai & Zhang 1993; Riley & Reynolds 1996). The IMR observations at Bourke showed that Australian plague locusts shifted their headings to the right when flying northwards and to the left when flying southwards. The exception to this occurred when locusts were flying towards the east when they undertook almost downwind flight. This orientation strategy has an adaptive advantage in eastern Australia, where the rainfall generally occurs in the temperate zone during winter and spring and in the subtropical zone in summer, while coastal regions receive rainfall all year round (Tapper & Hurry 1993). Shifting headings in relation to direction, the locusts would have avoided going too far into the arid inland where rainfall is scarce and also avoided migrating to cold, high latitude regions. Only successful migrants that reached areas with rich vegetation from rainfall or with a potential for rainfall would have survived and maximised the population’s reproductive output. During 1998-2001, C. terminifera developed to plague levels from very low background populations. The locust migrations detected by the IMRs suggest neither fixed directions nor definite routines have been adapted by the species. Successful migrations following rain distributions had enabled locust populations to increase markedly in every generation under favourable conditions within the agricultural regions and the arid inland between spring 1999 and spring 2000. The collapse of the locust outbreak in summer 2000-2001 also has proven that the locust plague had been significantly suppressed by the weather conditions after migrating into the arid inland where unusually little rain fell during that summer, in addition to the efforts of control campaigns. Since locusts migrate only in disturbed weather, i.e. tropical trough or col in eastern Australia, their orientation behaviour might have adapted to the weather patterns as neither the tropical trough nor

208 Chapter 6: Orientation of Australian Plague Locust the col will go too far into the inland Australia (Tapper & Hurry 1993). Strong winds could carry the locusts across long distances into potentially favourable habitats with rich vegetation after rain. Consequently, natural selection might have resulted the adaptive orientation behaviour from successful migrations.

Understanding the orientation scheme of plague locust will benefit the forecasting of population migration and redistribution and an adaptation to the unique environment of eastern Australia may be one explanation. Downwind orientation would allow the meteorological trajectory models used to simulate the locust migration routes (Deveson & Hunter 2002; Hunter & Deveson 2002). However, for accurate forecasting, caution must be taken in the interpretation of simulated trajectories, since the orientation might have taken the locusts significantly away from the modelled trajectories. Further study will be required to understand the mechanism of locust orientation. Observations of the infrequent southward migrations in autumn, from the inland into the agricultural zone, would be helpful in understanding both the locust orientation mechanisms and their adaptiveness for this species to a changing environment.

7 Discussion and Conclusions

1 FEASIBILITY OF REAL-TIME OBSERVATION...... 209 2 TARGET IDENTIFICATION FROM ECHO CHARACTERS...... 212 3 ESTIMATION OF MIGRATION PATH INCORPORATING ORIENTATION .... 214 4 EFFECTIVENESS FOR MONITORING LOCUST MOVEMENTS...... 215

The IMR mini-network in eastern Australia has demonstrated its application value in long-term monitoring of key migratory insects and intensive study of insect migration and flight behaviour. The fully automated operation and data processing system has enabled two IMRs to operate unattendedly in the remote interior. Target identification could rely solely on echo signatures, and the regional key pest, the Australian plague locust, can be identified and thus its migrations can be monitored in near real-time. Information on IMR detected migrations of plague locust can be used to inform decision making on the location and timing of field surveys, provide early warning of developments, forecast outbreaks and invasions and facilitate early control intervention. The accumulated data can also be used to improve understanding of locust migration and spatial population dynamics. The IMR mini-network has proven operationally reliable and practicable, and valuable in terms of furthering the understanding of insect migration and practical migratory pest management. The achievements from this study and possible improvements in future are discussed in this chapter.

1 Feasibility of Real-time Observation

Introduction of IMR with ZLC configuration was expected to simplify the operation of entomological radars in respect of long-term monitoring of insect migration and to

209 210 Chapter 7: Discussion and Conclusions deliver cost efficiencies by substantially reducing the requirement for intensive labour inputs and expensive running costs (Riley et al. 1992a; Drake 1993). This study has made a significant contribution on automating IMR operations for use in both migratory insect research and pest management. Practicable solutions have been achieved, by first digitising the radar signal and then analysing the signal to extract insect flight and identity parameters later after nightly radar observations (Drake et al. 2002a). The study also resulted in a significant improvement in the frequency of radar observations. For example, instead of the typical 5 min observation per half hour observation period (Smith et al. 2000), the IMRs in eastern Australian have increased the observation time up to 30-40 min and can be extend to 45 min for every hour still with radar hardware and performance checked hourly (Drake et al. 2002a). The IMRs operate in two alternative modes (Drake et al. 2002a), produce quality data on wingbeating from the stationary-beam observations (Drake et al. 2002b) in addition to data on target trajectory and body size and shape data from the rotary-beam observations (Harman & Drake 2004). Although it has been discovered recently that the wingbeating information can also be retrieved from some good-quality rotary-beam echoes (Wang & Drake 2004), the stationary-beam operation is considered valuable, especially for retrieval of wingbeating data from small insect targets. In terms of operational efficiency, the operation time was above 85% of scheduled time at Bourke during the three year period from June 1998 to May 2001 and 88% at Thargomindah from its first-year operation. The IMRs can operate all year around with extended maintenance intervals of 4 – 6 months.

The new algorithm of one-way peak-search for echo delimitation developed by this study has significantly improved echo quality and processing efficiency. Compared with the pre-set threshold method (Smith & Riley 1996; Drake et al. 1998; Smith et al. 2000), this dynamic approach has markedly reduced the affects of rain precipitation and nearby echo interference, and produced reliable results from observations under heavy migrations and light rains when virga may occur. The algorithm has demonstrated an ability to handle major migration events (for example, processing 54,000 targets for eight 50 m range gates of the Bourke IMR in a single night).

All software has been developed with C++ language, providing rapid processing and simplified data transformation. New algorithms of singular value decomposition for 1 Feasibility of IMR Real-time Observation 211 least-squares fit and fast Fourier transform for spectral analysis have been programmed and improved the computation efficiency and stability of data processing. The results from overnight IMR observations are usually available in graphic and text formats, also produced by self-developed software, within a few hours. This provides a near real-time monitoring system for insect migration. All software had been self-developed, which makes it easy the maintenance and further development.

An improved real-time IMR network could be achieved with some improvements including:

a) Hardware: The current IMR control and data processing system is based on a 486 PC, upgrade is inevitable. As advances of computer technology, ISA data acquisition cards will be discontinued and replaced by more popular and powerful PCI ones. This will make the replacement to current system difficult. Whilst the current PC is adequate for the task, a more modern PC, for example a PC with a Pentium IV CPU would substantially improve the speed of data processing. Using the current PC system, the longest time for data processing at Bourke was ~6 hr equivalent to approximately 2/3 of the IMR observation time. An upgraded PC system would reduce the processing time to a few minutes for an equivalent dataset meaning virtual real-time processing.

b) Software: The IMR software could also be improved to achieve real-time observation. With a fast PC, real-time signal processing could be implemented by incorporating the signal-processing program into the IMR-operating system, processing IMR signals during the observation intervals or at the same time when the digitised signals are still in the PC memory. The signal processing program still has room to improve. The wingbeating information can only be retrieved at ~20% from RB echoes with current DE software. The algorithm could be further improved to possibly retrieve wingbeat frequency from all good-quality RB echoes, which are about 35% of all target echoes.

c) Operating System: Migration of current single-tasking DOS to a multitasking OS such as Linux would enable the system to perform several tasks simultaneously. For example, on a Linux system, the signal processing procedure may not need to be incorporated into the IMR operation program. Instead, the signal processing could be executed separately but run at the same time as when the IMR is in observation. At the meantime, the networking is available which will allow to monitor the IMR operation

212 Chapter 7: Discussion and Conclusions and to access the observation results instantly. This will make the IMR observations available on the Internet in real-time if the PC is permanently connected to the Internet. Moreover, a multitasking OS like Linux will enable maximising usage of computer resources, which has always been a challenge under DOS, for example, the limits on conventional memory and hard disk size.

Transplant of all IMR software from DOS to another platform such as Linux is straightforward in terms of C++ coding. Online publishing the IMR observation in real- time requires, however, some precautions to be taken to ensure that, for example, heavy Internet traffic does not interrupt the IMR operation, slow down the data processing, or even crash the system.

2 Target Identification from Echo Characters

Target identification of Australian plague locust has been reliably achieved using echo signatures from the IMR observations in both rotary- and stationary-beam modes. Wingbeat frequency, however, has not been found as useful as expected; using the wingbeat frequency alone cannot identify detected targets into species (Schaefer 1969; Riley 1974; Schaefer 1976; Riley & Reynolds 1979). The harmonic character of wing beating, on the other hand, could be valuable to identify IMR targets. The target size and shape parameters, especially the shape coefficients, are powerful parameters in separating target species as target sizes can overlap largely among species and spread widely between individuals of the same species but the shape factors are relatively unique. The polarisation-averaged RCS term a0 has been found a good estimator for target size and an empirical formula has been developed to estimate a target size from the RCS area into mass measure. The ratios of aa20 and aa40 have been found to exhibit particular distribution patterns and peak values that can be used, in addition to size, to better identify targets.

Wingbeat frequency has been retrieved successfully from IMR samples under rotary-beam observation mode (Wang & Drake 2004). Although wingbeat frequency has been confirmed to be dependent on temperature, the direct link to insect size and shape and the lapse rate with temperature provides additional information for target 2 Target Identification from Echo Characters 213 identification. The relationship of wingbeat frequency to insect size needs further investigation and study using a large sample.

The echo characters of other key migratory insects in eastern Australia could be identified using the same method for Australian plague locust. For example, the native budworm moth, Helicoverpa punctigera, often migrates in the interior regions of eastern Australia between July and October (Fitt et al. 1995). These moths are similar in size to the plague locusts but have a quite different shape. Collecting more ancillary information may enable the moths to be discriminated from the detected migrations from archived IMR data. Other direct evidence of population movement, such as field survey, could be used when a major migration is detected by the IMR. With this indirect calibration method, a library of echo signatures for key migrants can be built up and fine-tuned, and thus the IMR could be used to monitor additional species.

A systematic approach to identification of IMR targets from signal will improve the reliability. Owing to the lack of laboratory facilities and aerial sampling equipment, and the infeasibility of deploying aerial sampling equipment to the remote IMR sites in real time, the radar signatures of C. terminifera were identified from the IMR detected populations which from ancillary information were considered most likely to be the Australian plague locusts. By using these species criteria, the plague locust migrations were identified from the IMR observations and the general success of reconstruction of their migration routes seems to be in good agreement with the development of field populations and the persistence of a metapopulation in eastern Australia. It would be ideal that aerial samples are taken during the IMR operation (Chapman et al. 2004). Thus the migrating fauna can be identified for the specific location and time and detailed measures on their geometric sizes and radar characters of captured specimen can be measured with a laboratory system (Riley 1985; Chapman et al. 2002b). These aerial specimens are also valuable for comparative studies in association with the field populations. Therefore understanding physiological aspects of their migration, such as the adult age, the status of reproductive development and the flight fuel reserve and usage, can be improved. With the categorised echo signatures, similar migrating populations can be discriminated from other airborne species. If a laboratory facility is available for measuring the radar scattering terms of insect specimen, field samples of the young adults of migratory species can be collected and measured. Then these

214 Chapter 7: Discussion and Conclusions measures can be used as a key to match IMR observed migrating insects. However, this approach may require a deep background research. Further study on the radar scattering characters in association with insect geometric features and the wingbeat frequency and its harmonic patterns with temperature dependence may provide additional insights useful for IMR target identification (Dean & Drake 2005).

3 Estimation of Migration Path Incorporating Orientation

Trajectory analysis has been a very useful approach to estimate the direction and distance of insect migrations (Rosenberg & Magor 1987). The downwind migration in most migratory insects enables an air particle to be used as a model to simulate the insects’ migration trajectories in the wind field. The flight parameters measured by the IMR can be used alone to calculate trajectories on the assumption that migrating insects some distance away made the same course changes, at the same time, as those flying over the IMR. Although migration trajectories can be estimated from IMR observations directly, understanding orientation behaviours helps forecast plague locust migrations, since migrating plague locusts have invariably been detected with a strong collective orientation: downwind but with varied offset angles. The assumption that the migrating individuals follow the similar trend to those detected by the IMR may not be always true. For example, within a nearly stationary trough system, migrating insects detected by the IMR under the trough line can hardly be simulated either by a pure wind trajectory model or by extrapolation from the time-series of IMR observations: the airspeeds and orientation angles of the insects and the wind speeds and directions within the trough system change too rapidly. From the three-year study period of 1998 – 2001, it is clear that some significant locust migrations may have been missed by the IMRs due to their incomplete coverage across eastern Australia: for example, the late southward migrations from the interior during the season of 1999 – 2000 and the possible early northward migrations from the Riverina of NSW during the season of 2000 – 2001. Meteorological trajectory models, in these cases, can be useful in simulating these migration paths based on the available evidence from other sources such as locust field surveys and light trap data. However, without the consideration of locust orientation, the migration paths estimated from wind trajectory models could be 3 Estimation of Migration paths Incorporating Orientation 215 subject to substantial error of up to several hundreds of kilometres in respect of the actual origin and destination regions for a migration event extending over 2-3 successive nights. By taking orientation behaviour into account, migration paths might be estimated from meteorological data with a higher accuracy. The airspeed and crosswind angle of plague locusts can be combined vectorially with simulated winds to calculate modified wind trajectories. Nevertheless, the mechanism of plague locust orientation is not, as yet, fully understood. They migrate northwards with right-shifted headings but southwards with left-shifted headings. Therefore using a modified wind trajectory model, including orientation parameters, to estimate locust migrations will need further study.

4 Effectiveness for Monitoring Locust Movements

IMR has demonstrated its application and value in both the study of insect migration and more generally in pest management. With this study, the understanding of plague locust migrations in eastern Australia has been improved. The finding that the arid interior is not invariably the initial source (Wright 1987) of a major locust outbreak or plague and that northward migrations play an important role in the population build-up in the inland area (Deveson et al. 2005) has important strategic implications for locust monitoring and management. The overwintering populations in the south of NSW and the north of Vic would be critical in the seasonal dynamics of the locust metapopulation in a normal season when the winter rain falls in the south and the summer rain falls in the interior of eastern Australia. The egg-laying activity in autumn and the hatching rate in spring in this overwintering region thus provide an indication whether the locusts could produce an outbreak in the coming season. Therefore, an early warning could be issued after close monitoring of this region. Whether the eggs laid in the arid to semi- arid inland can survive the normally dry winter there and produce a rapid population build-up from abnormal winter rainfall in this subtropical region remains subject for future study.

The effectiveness of the IMR mini-network for monitoring the movements of key migratory insects between the interior and the southeastern agricultural region has

216 Chapter 7: Discussion and Conclusions been partially demonstrated. As noted, the IMRs only detected a few southward migrations of plague locust during the study period; the dominant movements were the so-called return migrations into the interior. This was surprisingly different from the original concept that the IMRs were to monitor the population movements in the source and between the source and the infestation area (Drake et al. 2001). In eastern Australia, the migrations of plague locusts occur during later spring to autumn. Weather systems, as indicated by the latitudes of the centres of high pressures, move southwards in the spring and northwards in later autumn. The plague locusts migrate only in association with disturbed weather, either tropical troughs penetrating southwards commonly in spring and summer or pre-frontal temperate troughs associated with cold fronts in autumn. As usually the cold fronts become weak before approaching the Great Dividing Range, the pre-frontal troughs do not extend into the eastern inland. Therefore, the southward migrations would occur between the Far North and Northeast SA and the Far Southwest NSW and Northwest Vic; both IMRs are far away from this route. Nevertheless, this finding enriched the knowledge about the plague locust migration and changed the view that the plague locusts only conduct one-way migration from the interior to the agricultural region.

The current IMR network in eastern Australian comprises only two IMR units. This limits the effectiveness in terms on monitoring potentially important plague locust migrations and population re-distributions in the large area of eastern Australia within which plague locust populations can occur. The limited coverage of the area of interest will mean that some potentially important locust migration events will not be detected. The addition of one more IMR units or relocating current IMR units would provide more effective monitoring of locust movements between the arid inland and the agricultural zone in eastern Australia. The current study has revealed that both IMR units are not on the most frequent migration paths of the locusts: the Bourke IMR cannot detect the locust migrations between the agricultural regions in the Riverina and the inland area; the Thargomindah IMR does not sit in the centre of locust habitats in the arid inland area. Adding an additional IMR unit located in the Far Southwest NSW, for example at Broken Hill, or relocating the Bourke IMR to Broken Hill and the Thargomindah IMR to Windorah, would improve the radar coverage of locust migration paths. 4 Effectiveness of Monitoring Locust Movements 217

From this study, the IMR has demonstrated its capability and utility for long- term migration monitoring of key migratory species, in addition to its usage as a research tool. During the study period, a large amount of data has been accumulated. The data could be used to study other key migratory insects and airborne insect fauna and diversity, as well as providing some directions for future plague locust research. With further development, the IMR network can be used as a real-time monitoring system on insect migration.

8 Bibliography

Aldhous, AC (1989). An Investigation of the Polarisation Dependence of Insect Radar Cross Sections at Constant Aspect. PhD Thesis. Ecological Physics Research Group, Cranfield Institute of Technology. Andria, G, M Savino and A Trotta (1993). FFT-based algorithms oriented to measurements on multifrequency signals. Measurement 12(1), 25-42. Applebaum, SW, E Avisar and Y Heifetz (1998). Juvenile hormone and locust phase. Archives of Insect Biochemistry and Physiology 35(4), 375-391. Atlas, D and FI Harris (1970). Measurement of point target speeds with incoherent non-tracking radar: insect speeds in atmospheric waves. Journal of Geophysical Research 75(36), 7588- 7595. Atlas, D, JI Metcalf, JH Richter and EE Gossard (1970). The birth of "CAT" and microscale turbulence. Journal of the Atmospheric Sciences 26, 903-913. Australian Government Bureau of Meteorology (2004). Australia's Climate. http://www.abs.gov.au/ausstats/[email protected]/0/35f9cdf3307c036eca256f7200832f12?OpenDoc ument. Visited on 24 June 2005. Baas, JH (2000). EZ-ROSE: a computer program for equal-area circular histograms and statistical analysis of two-dimensional vectorial data. Computers & Geosciences 26, 153- 166. Barndorff-Nielsen, O and DR Cox (1979). Edgeworth and saddle-point approximations with statistical applications (with discussion). Journal of the Royal Statistical Society. Series B (Methodological) 41, 279-312. Batschelet, E (1981). Circular Statistics in Biology. London: Academic Press. Beerwinkle, KR, JD Lopez and JA Witz (1988). Emergence and radar-observed movement of Heliothis zea from senescent corn in east-central Texas. In Beltwide Cotton Production Res. Conf., pp. 329-332. Beerwinkle, KR, JA Witz and PG Schleider (1993). An automated, vertical looking, X-band radar system for continuously monitoring aerial insect activity. Transactions of the ASAE 36, 965-970. Beerwinkle, KR, JD Lopez, Jr., PG Schleider and PD Lingren (1995). Annual patterns of aerial insect densities at altitude from 500 to 2400 meters in east-central Texas indicated by continuously-operating vertically-oriented radar. Southwestern Entomologist Supp18, 63- 79. Beerwinkle, KR, JD Lopez, Jr., JA Witz, PG Schleider, RS Eyster and PD Lingren (1994). Seasonal radar and meteorological observations associated with nocturnal insect flight at altitudes to 900 meters. Environmental Entomology 23(3), 676-683.

219 220 Bibliography

Beeston, G, G Spencer, A Howcroft and G Pickles (2001). New technology used to fight locusts. In Geospatial Information & Agriculture Conference: Incorporating Precision Agriculture in Australasia 5th Annual Symposium, Eveleigh, Sydney, 16-19 July 2001, pp. 375-381. Adelaide: Causal Productions Pty Ltd. Bent, GA (1984). Developments in detection of airborne aphids with radar. In British Crop Protection Conference - Pests and Diseases, Brighton Metropole, England, 19-22 November 1984, pp. 665-674. Croydon: British Crop Protection Council. Bent, GA, N Carter, R Harrington, GM Tatchell, MS Taylor and IP Woiwod (1987). The present status and future developments in the Rothamsted Insect Survey. In Aphid Migration and Forecasting-'Euraphid' Systems in European Community Countries, pp. 201- 207. Luxembourg: Commission of the European Communities. Best, DJ and NI Fisher (1981). The bias of the maximum likelihood estimators of the von Mises-Fisher concentration parameters. Commun. Statist. - Simul. Comput. B10(5), 493- 502. Bingman, VP and K Cheng (2005). Mechanisms of animal global navigation: comparative perspectives and enduring challenges. Ethology Ecology & Evolution 17(4), 295-318. Blake, LV (1986). Radar Range-Performance Analysis. Norwood, Ma.: Artech House, Inc. Brady, J (1997). J. S. Kennedy (1912-1993): a clear thinker in behavior's confused world. Annual Review of Entomology 42, 1-22. Breitenbach, A (1999). Against spectral leakage. Measurement 25(2), 135-142. Burt, PJA (1998). Suction trap catches of airborne insects in relation to convection in the lower atmosphere. Aerobiologia 14, 255-266. Canavoso, LE, ZE Jouni, KJ Karnas, JE Pennington and MA Wells (2001). Fat metabolism in insects. Annu. Rev. Nutr. 21, 23-46. Capaldi, EA, AD Smith, JL Osborne, SE Fahrbach, SM Farris, DR Reynolds, AS Edwards, A Martin, GE Robinson, GM Poppy and JR Riley (2000). Ontogeny of orientation flight in the honeybee revealed by harmonic radar. Nature 403, 537-540. Chapman, JW, DR Reynolds and AD Smith (2003). Vertical-looking radar: A new tool for monitoring high-altitude insect migration. BioScience 53(5), 503-511. Chapman, JW, AD Smith, IP Woiwod, DR Reynolds and JR Riley (2002a). Development of vertical-looking radar technology for monitoring insect migration. Computers and Electronics in Agriculture 35(2-3), 95-110. Chapman, JW, DR Reynolds, AD Smith, ET Smith and IP Woiwod (2004). An aerial netting study of insects migrating at high altitude over England. Bulletin of Entomological Research 94(2), 123-136. Chapman, JW, DR Reynolds, SJ Brooks, AD Smith and IP Woiwod (2006). Seasonal variation in the migration strategies of the green lacewing Chrysoperla carnea species complex. Ecological Entomology 31, 378-388. Chapman, JW, DR Reynolds, AD Smith, JR Riley, DE Pedgley and IP Woiwod (2002b). High- altitude migration of the diamondback moth Plutella xylostella to U.K.: a study using radar, aerial netting, and ground trapping. Ecological Entomology 27, 641-650. Chapman, JW, DR Reynolds, AD Smith, JR Riley, MG Telfer and IP Woiwod (2005). Mass aerial migration in the carabid beetle Notiophilus biguttatus. Ecological Entomology 30, 264-272. Bibliography 221

Chapman, JW, DR Reynolds, H Mouritsen, JK Hill, JR Riley, D Sivell, AD Smith and IP Woiwod (2008). Wind selection and drift compensation optimize migratory pathways in a high-flying moth. Current Biology 18, 514-518. Chen, RL, XZ Bao, SY Wang, YJ Sun, LQ Li and JR Liu (1988). A radar equipment for entomological observation and its test. Acta Ecologica Sinica 8, 176-182. Chen, RL, XZ Bao, SY Wang, YJ Sun, LQ Li and JR Liu (1992). An observation on the migration of meadow moth by radar. Acta Phytophylacica Sinica 19(2), 171-174. Chen, RL, XZ Bao, VA Drake, RA Farrow, SY Wang, YJ Sun and BP Zhai (1989). Radar observations of the spring migration into northeastern China of the oriental armyworm moth, Mythimna separata, and other insects. Ecological Entomology 14, 149-162. Chen, RL, XZ Bao, SY Wang, YJ Sun, LQ Li, JR Liu, DK Zhang and J Lu (1985). The Gongzhuling entomological radar and observations of the migrations of armworm and meadow moths. Scientia Agricultura Sinica(3), 93. Cheng, DF, KM Wu, Z Tian, LP Wen and ZR Shen (2002). Acquisition and analysis of migration data from the digitised display of a scanning entomological radar. Computers and Electronics in Agriculture 35(2-3), 63-75. Cheng, XN, XX Zhang, JY Cheng, JR Riley and DR Reynolds (1994). Radar observations of the autumn migration of Nilaparvata lugens (Homoptera: Delphacidae) in the East of China. Journal of Nanjing Agricultural University 17(3), 24-32. Cheng, XN, RC Chen, X Xi, LM Young, ZL Zhu, JC Wu, RG Qian and JS Young (1979). Studies on the migrations of brown planthopper Nilaparvata lugens (Stål). Acta Entomologica Sinica 22(1), 1-21. Cherlet, MR, A Di Gregorio and JU Hielkema (1991). Remote-sensing applications for desert- locust monitoring and forecasting. Bulletin OEPP/EPPO Bulletin 21, 633-642. Clark, DP (1969). Night flights of the Australian plauge locust, Chortoicetes terminifera Walk., in relation to storms. Australian Journal of Zoology 17, 329-352. Clark, DP (1971). Flights after sunset by the Australian plague locust, Chortoicetes terminifera (Walk.), and their significance in dispersal and migration. Australian Journal of Zoology 19, 159-176. Clark, DP, C Ashall, Z Waloff and L Chinnick (1969). Field studies on the Australian plague locust (Chortoicetes terminifera Walk.) in the 'Channel Country' of Queensland. Anti- Locust Bulletin 44, 101 pp. London: Anti-Locust Research Centre. Common, IFB (1952). Migration and gregarious aestivation in the Bogong moth, Agrotis infusa. Nature 170(4336), 981-982. Common, IFB (1954). A study of the ecology of the adult bogong moth, Agrotis infusa (Boisd.) (Lepidoptera: Noctuidae), with special reference to its behaviour during migration. Australian Journal of Zoology 2(2), 223-263. Cressman, K (1997). SWARMS: A geographic information system for Desert Locust forecasting. In New Strategies in Locust Control, S Krall, R Peveling and D Ba Diallo (eds), pp. 27-35. Basel, Switzerland: Birkhäuser Verlag. Danthanarayana, W (1986). Insect Flight: Dispersal and Migration. Berlin: Springer-Verlag. Dean, TJ and VA Drake (2005). Monitoring insect migration with radar: the ventral-aspect polarization pattern and its potential for target identification. International Journal of Remote Sensing 26(18), 3957-3974.

222 Bibliography

Deng, WX (1981). A general survey on seasonal migrations of Nilaparvata lugens Stal and Sogatella furcifera (Horrath) (Homoptera: Delphacidae) by means of airplane collection. Acta Phytophylacica Sinica 8(2), 73-83. Dent, D (2000). Insect Pest Management. Oxon, UK: CABI Publishing. Deveson, ED (2001). Decision support for locust management using GIS to integrate multiple information sources. In Geospatial Information & Agriculture Conference: Incorporating Precision Agriculture in Australasia 5th Annual Symposium, Eveleigh, Sydney, 16-19 July 2001, pp. 361-374. Adelaide: Causal Productions Pty Ltd. Deveson, ED and DM Hunter (2002). The operation of a GIS-based decision support system for Australian locust management. Entomologia Sinica 9(4), 1-12. Deveson, ED and PW Walker (2005). Not an one-way trip: historical distribution data for Australian plague locusts support frequent seasonal exchange migrations. Journal of Orthoptera Research 14(1), 91-105. Deveson, ED, VA Drake, DM Hunter, PW Walker and HK Wang (2005). Evidence from traditional and new technologies for northward migrations of Australian plague locusts (Chortoicetes terminifera) (Walker) (Orthoptera: Acrididae) to western Queensland. Australian Ecology 30, 920-935. Dickson, RBBB (1990). Detection of mesoscale synoptic features associated with dispersal of spruce budworm moths in eastern Canada. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 328(1251), 607-617. Dickson, RBBB, MJ Haggis, RC Rainey and LMD Burns (1986). Spruce budworm moth flight and storms: futher stuides using aircraft and radar. Journal of Climate and Applied Meteorology 25, 1600-1608. Dillon, ML, GP Fitt, JG Hamilton and WA Rochester (1996). A simulation model of wind- driven dispersal of Helicoverpa moths. Ecological Modelling 86, 145-150. Dingle, H (1996). Migration: the Biology of Life on the Move. New York: Oxford University Press. Dingle, H (2001). The evolution of migratory syndromes in insects. In Insect Movement: Mechanisms and Consequences, I Woiwood, DR Reynolds and CD Thomas (eds), pp. 159- 181. Oxon: CABI Publishing. Dingle, H and R Winchell (1997). Juvenile hormone as a mediator of plasticity in insect life histories. Archives of Insect Biochemistry and Physiology 35(4), 359-373. Dingle, H and VA Drake (2007). What is migration? BioScience 57(2), 113-121. Dixon, AFG, S Horth and P Kindlmann (1993). Migration in insects: cost and strategies. The Journal of Animal Ecology 62(1), 182-190. Donadio, M (1999). How to Interpolate Frequency Peaks. Last updated on 05 May 1999. http://dspguru.com/howto/tech/peakfft.htm. Visited on 15 Mar 2001. Downing, JD and EL Frost (1972). Recent radar observations of diurnal insect behavior. In Fifty-Ninth Annual Meeting, New Jersey Mosquito Extermination Association, pp. 114-132. Drake, VA (1981a). Quantitative Observation and Analysis Procedures for a Manually Operated Entomological Radar. Technical Paper 19, 41 pp. Melbourne: Division of Entomology, CSIRO. Drake, VA (1981b). Target density estimation in radar biology. J. theor. Biol. 90, 545-571. Drake, VA (1982a). Insects in the sea-breeze front at Canberra: a radar study. Weather 37(5), 134-143. Bibliography 223

Drake, VA (1982b). The CSIRO entomological radar: a remote-sensing instrument for insect migration research. In Scientific Instruments in Primary Production, LA Wisbey (ed), pp. 63-73. Melbourne: Australian Scientific Industry Association. Drake, VA (1983). Collective orientation by nocturnally migrating Australian plague locusts, Chortoicetes terminifera (Walker) (Orthoptera: Acrididae): a radar study. Bulletin of Entomological Research 73(4), 679-692. Drake, VA (1984a). The vertical distribution of macro-insects migrating in the nocturnal boundary layer: a radar study. Boundary Layer Meteorology 28(3-4), 353-374. Drake, VA (1984b). A solitary wave disturbance of the marine boundary layer over Spencer Gulf revealed by radar observations of migrating insects. Aust. Met. Mag. 32, 131-135. Drake, VA (1985a). Solitary wave disturbances of the nocturnal boundary layer revealed by radar observations of migrating insects. Boundary Layer Meteorology 31, 269-286. Drake, VA (1985b). Radar observations of moths migrating in a nocturnal low-level jet. Ecological Entomology 10, 259-265. Drake, VA (1990a). The "Scanning Radar Microscope": a radar configuration for microscale meteorological research. In Radarcon 90, Adelaide, Australia, 18-20 April 1990, DJ Heilbronn (ed), pp. 603-609. Defence Science and Technology Organisation. Drake, VA (1990b). Methods for studying adult movement in Heliothis. In Heliothis: Research Methods and Prospects, MP Zalucki (ed), pp. 109-121. New York: Springer-Verlag. Drake, VA (1991). The influence of weather and climate on insect pests and insect-borne diseases: an Australian perspective. In Agricultural Meteorology, Melbourne, 17-19 July 1991, pp. 1-8. Drake, VA (1993). Insect-monitoring radar: a new source of information for migration research and operational pest forecasting. In Pest Control and Sustainable Agriculture, SA Corey, DJ Dall and WM Milne (eds), pp. 452-455. Melbourne: CSIRO Publications. Drake, VA (1994). The influence of weather and climate on agriculturally important insects: an Australian perspective. Australian Journal of Agricultural Research 45(3), 487-509. Drake, VA (1998). Forecasting highly mobile pests - confronting the limits of predictability? In Pest Management - Future Challenges. Proceedings of the Sixth Australasian Applied Entomological Research Conference, Brisbane, Australia, 28 September - 2 October 1998, MP Zalucki, RAI Drew and GG White (eds), 2, pp. 63-72. The University of Queensland Printery. Drake, VA (2001). The Radar Entomology Web Site. Last updated on 04 Dec 2001. http://www.ph.adfa.edu.au/a-drake/trews/. Visited on 18 Apr 2002. Drake, VA (2002). Automatically operating radars for monitoring insect pest migrations. Entomologia Sinica 9(4), 27-39. Drake, VA and RA Farrow (1983). The nocturnal migration of the Australian plague locust, Chortoicetes terminifera (Walker) (Orthoptera: Acrididae): quantitative radar observations of a series of northward flights. Bulletin of Entomological Research 73, 567-585. Drake, VA and RA Farrow (1985). A radar and aerial-trapping study of an early spring migration of moths (Lepidoptera) in inland New South Walkes. Australian Journal of Ecology 10, 223-235. Drake, VA and RA Farrow (1988). The influence of atmospheric structure and motions on insect migration. Annual Review of Entomology 33, 183-210. Drake, VA and RA Farrow (1989). The 'Aerial Plankton' and atmospheric convergence. Trends in Ecology and Evolution 4(12), 381-385.

224 Bibliography

Drake, VA and AG Gatehouse (1995). Insect Migration: Tracking Resources through Space and Time. Cambridge: Cambridge University Press. Drake, VA, IT Harman and DM Hunter (1998). Monitoring locust migratory flight with radar: an Australian field study. In Second Urban Environment Symposium and 13th Conference on Biometeorology and Aerobiology, Albuquerque, New Mexico, 2-6 November 1998, pp. 313-316. Boston: American Meteorological Society. Drake, VA, IT Harman and HK Wang (2000). Automatically operating radars for monitoring insect pest migrations. In XXI International Congress of Entomology, Foz do Iguassu, Brazil, 20-26 August 2000, DL Gazzoni (ed). Londrina: Embrapa Soja. Drake, VA, HK Wang and IT Harman (2002a). Insect monitoring radar: remote and network operation. Computers and Electronics in Agriculture 35(2-3), 77-94. Drake, VA, IT Harman and HK Wang (2002b). Insect monitoring radar: stationary-beam operating mode. Computers and Electronics in Agriculture 35(2-3), 111-137. Drake, VA, KF Helm, JL Readshaw and DG Reid (1981). Insect migration across Bass Strait during spring: a radar study. Bull. ent. Res. 71, 449-466. Drake, VA, IT Harman, IA Bourne and GJC Woods (1994). Simultaneous entomological and atmospheric profiling: a novel technique for studying the biometeorology of insect migration. In 21st Conf. Agric. For. Meteorol. & 11th Conf. Biomet. Aerobiol., San Diego, California, 7-11 March 1994, pp. 444-447. Boston: American Meteorological Society. Drake, VA, PC Gregg, IT Harman, HK Wang, ED Deveson, DM Hunter and WA Rochester (2001). Characterizing insect migration systems in inland Australia with novel and traditional methodologies. In Insect Movement: Mechanisms and Consequences, I Woiwood, DR Reynolds and CD Thomas (eds), pp. 207-233. Oxon: CABI Publishing. Dudley, R (2000a). The Biomechanics of Insect Flight: Form, Function, Evolution. Princeton, New Jersey: Princeton University Press. Dudley, R (2000b). The evolutionary physiology of animal flight: paleobiological and present perspectives. Annual Review of Physiology 62, 135-155. Dudley, R (2001). The biomechanics and functional diversity of flight. In Insect Movement: Mechanisms and Consequences, I Woiwood, DR Reynolds and CD Thomas (eds), pp. 19- 41. Oxon: CABI Publishing. Fairbairn, DJ (1994). Wing dimorphism and the migratory syndrome: Correlated traits for migratory tendency in wing dimorphic insects. Researches on Population Ecology 36(2), 157-163. Farrow, RA (1977). Origin and decline of the 1973 plague locust outbreak in central western New South Wales. Australian Journal of Zoology 25, 455-489. Farrow, RA (1979). Population dynamics of the Australian plague locust, Chortoicetes terminifera (Walker), in central western New South Wales. I. Reproduction and migration in relation to weather. Australian Journal of Zoology 27, 717-745. Farrow, RA (1981a). Aerial dispersal of Scelio fulgidus [Hym.: Scelionidae], parasite of eggs of locusts and grasshoppers[Ort.: Acrididae]. Entomophaga 26(4), 349-355. Farrow, RA (1981b). Aerial dispersal of microinsects. In 3rd Australiasian Conf. Grassl. Invert. Ecol., Adelaide, 30 Nov - 4 Dec 1981, pp. 51-55. Farrow, RA (1986). Interactions between synoptic scale and boundary-layer meteorology on micro-insect migration. In Insect Flight: Dispersal and Migration, W Danthanarayana (ed), pp. 185-195. Berlin: Springer-Verlag. Bibliography 225

Farrow, RA (1991). Migration strategies of insect pests of agriculture: a review of mechanisms. In International Seminar on Migration and Dispersal of Agricultural Pests, Tsukuba, Japan, 25-28 September 1991, pp. 1-19. Farrow, RA and JE Dowse (1984). Method of using kites to carry tow nets in the upper air for sampling migrating insects and its application to radar entomology. Bull. ent. Res. 74, 87- 95. Farrow, RA and G McDonald (1987). Migration strategies and outbreaks of noctuid pests in Australia. Insect Science and its Application 8(4-6), 531-542. Farrow, RA and JC Daly (1987). Long-range movements as an adaptive strategy in the genus Heliothis (Lepidoptera:Noctuidae): a review of its occurrence and detection in four pest species. Aust. J. Zool. 35(1), 1-24. Farrow, RA and VA Drake (1991). Migration. In The Insects of Australia: a Textbook for Students and Research Workers, CSIRO (ed), pp. 103-108. Carlton: Melbourne University Press. Feng, CH, BP Zhai and XX Zhang (2001). Re-emigration capacity of the brown planthopper, Nilaparvata lugens. Chinese Journal of Rice Sciences 15(2), 125-130. Feng, HQ, KM Wu, DF Cheng and YY Guo (2003). Radar observations of the autumn migration of the beet armyworm Spodoptera exigua (Lepidoptera: Noctuidae) and other moths in northern China. Bulletin of Entomological Research 93, 115-124. Feng, HQ, KM Wu, YX Ni, DF Cheng and YY Guo (2006). Nocturnal migration of dragonflies over the Bohai Sea in northern China. Ecological Entomology 31, 511-520. Feng, HQ, YH Zhang, KM Wu, DF Cheng and YY Guo (2007). Nocturnal windborne migration of ground beetles, particularly Pseudoophonus griseus (Coleoptera: Carabidae), in China. Agricultural and Forest Entomology 9, 103-113. Fisher, NI (1993). Statistical Analysis of Circular Data. Cambridge: Cambridge University Press. Fitt, GP, ML Dillon and JG Hamilton (1995). Spatial dynamics of Helicoverpa populations in Australia: simulation modelling and empirical studies of adult movement. Computers and Electronics in Agriculture 13(2), 177-192. Flanagan, D (1998). JavaScript: The Definitive Guide: O'Reilly & Associates, Inc. Frost, EL (1971a). Correlated radar and trapping studies of insect swarming over atlantic coastal lowlands. In 7th Symposium on Remote Sensing of the Environment III, pp. 1905-1915. Frost, EL (1971b). Radar returns and correlated insect trapping data. In Fifty-Eighth Annual Meeting, New Jersey Mosquito Extermination Association, pp. 81-92. Frost, EL and J Robinson (1973). Radar entomology on the great plains. In Sixtieth Annual Meeting, New Jersey Mosquito Extermination Assoication, pp. 160-169. Gagnon, R (1999). Java and JavaScript HowTo. http://www.rgagnon.com/howto.html. Visited on 24/05/2000. Gatehouse, AG (1997). Behavior and ecological genetics of wind-borne migration by insects. Annual Review of Entomology 42, 475-502. Gentle, JE (1998). Numerical Linear Algebra for Applications in Statistics. New York: Springer-Verlag. Geoscience_Australia (2002). Map and Images of Australia. Last updated on 3 July 2002. http://www.agso.gov.au/map/. Visited on 4 July 2002.

226 Bibliography

Gibo, DL (1986). Flight strategies of migrating monarch butterflies (Danaus plexippus L.) in southern Ontario. In Insect Flight: Dispersal and Migration, W Danthanarayana (ed), pp. 172-184. Berlin: Springer-Verlad. Glover, KM and KR Hardy (1966). Dot angels: insects and birds. In 12th Weather Radar Conference, pp. 264-268. Boston, Mass: American Meteorological Society. Glover, KM, KR Hardy, CR Landry and TG Konrad (1966a). Radar characteristics of know insects in free flight. In 12th Weather Radar Conference. Boston, Mass: American Meteorological Society. Glover, KM, KR Hardy, TG Konrad, WN Sullivan and AS Michaels (1966b). Radar observations of insects in free flight. Science 154(3752), 967-972. Golub, GH and C Reinsch (1971). Singular value decomposition and least squares solutions. In Handbook for Automatic Computation, Volume II. Linear Algebra, JH Wilkinson and C Reinsch (eds). New York: Springer-Verlag. Gossard, EE and RB Chadwick (1979). Studies of insect by high resolution radar. In 14th Conf. of Agriculture and Forest Meteorology and 4th Conf. on Biometeorology, Boston, MA, pp. 268-271. Amer. Met. Soc. Goto, Y (2000). Interpolation of Hamming-apodized DFT spectra. Electronics & Communications in Japan 83(6), 107-115. Greenbank, DO, GW Schaefer and RC Rainey (1980). Spruce budworm (Lepidoptera: Tortricidae) moth flight and dispersal: new understanding from canopy observations, radar, and aircraft. Memoirs of the Entomological Society of Canada 110, 49 pp. Gregg, PC, AP Del Socorro and WA Rochester (2001). Field test of a model of migration of moths (Lepidoptera: Noctuidae) in inland Australia. Australian Journal of Entomology 40, 249-256. Gregg, PC, GP Fitt, M Coombs and GS Henderson (1994). Migrating moths collected in tower- mounted light traps in northern New South Wales, Australia. Bulletin of Entomological Research 84(1), 17-30. Gregg, PC, GP Fitt, MP Zalucki and DAH Murray (1995). Insect migration in an arid continent. II. Helicoverpa spp. in eastern Australia. In Insect Migration: Tracking Resources through Space and Time, VA Drake and AG Gatehouse (eds), pp. 151-172. Cambridge: Cambridge University Press. Gunn, A and AG Gatehouse (1993). The migration syndrome in the African armyworm moth, Spodoptera exempts: allocation of resources to flight and reproduction. Physiological Entomology 18(2), 149-159. Gunn, A, AG Gatehouse and KP Woodrow (1989). Trade-off between flight and reproduction in the African armyworm moth, Spodoptera exempta. Physiological Entomology 14(4), 419- 427. Hagler, JR and CG Jackson (2001). Methods for marking insects: current techniques and future prospects. Annual Review of Entomology 46, 511-543. Hajovsky, RG, AP Deam and AH Lagrone (1966). Radar reflections from insects in the lower atmosphere. IEEE Transactions on Antennas and Propagation 14(2), 224-227. Hamilton, JG (1998). Forecasting and locust management or why some insects aren't locusts. In Pest management - Future challenges. Proceedings of the Sixth Australasian Applied Entomological Research Conference, Brisbane, Australia, 28 September - 2 October 1998, MP Zalucki, RAI Drew and GG White (eds), 2, pp. 82-86. The University of Queensland Printery. Bibliography 227

Hamilton, JG, WA Rochester and PC Gregg (1994). Predicting long-distance migration of insect pests in eastern Australia. In 21st Conf. Agric. For. Meteorol. & 11th Conf. Biomet. Aerobiol., San Diego, California, 7-11 March 1994, pp. 431-434. Boston: American Meteorological Society. Hardy, KR and KM Glover (1966). 24 hour history of radar angel activity at three wavelengths. In 12th Weather Radar Conference, pp. 269-274. Boston, Mass: American Meteorological Society. Hardy, KR and I Katz (1969). Probing the clear atmosphere with high power, high resolution radars. Proceedings of the IEEE, 468-480. Hardy, KR and H Ottersten (1969). Radar investigations of convective patterns in the clear atmosphere. Journal of the Atmospheric Sciences 26, 666-672. Hardy, KR, D Atlas and KM Glover (1966). Multiwavelength backscatter from the clear atmosphere. Journal of Geophysical Research 71(6), 1537-1552. Harman, IT and VA Drake (2004). Insect monitoring radar: analytical time-domain algorithm for retrieving trajectory and target parameters. Computers and Electronics in Agriculture 43(1), 23-41. Harris, FI (1978). On the use of windows for harmonic analysis with the discrete Fourier transform. Proceedings of the IEEE 66(1), 1-12. Healey, RG, SG Robertson, JI Magor, J Pender and K Cressman (1996). A GIS for Desert Locust forecasting and monitoring. International Journal of Geographical Information Systems 10(1), 117-136. Hendrie, LK, ME Irwin, NJ Liquido, WG Ruesin, EA Mueller, DJ Voegtlin, GL Achtemeier, WM Steiner and RW Scott (1985). Conceptual approach to modeling aphid migration. In The Movement and Dispersal of Agriculturally Important Biotic Agents, DR MacKenzie, CS Barfield, GG Kennedy, RD Berger and DJ Taranto (eds), pp. 541-582. Baton Rouge, Louisiana: Claitor's Publishing Division. Herman, WS and M Tatar (2001). Juvenile hormone regulation of longevity in the migratory monoarch butterfly. In Proceedings of the Royal Society of London - Biological Sciences, London, Dec 22, 2001, 268, pp. 2509-2514. The Royal Society. Hobbs, SE (1991). A radar signal processor for biological applications. Meas. Sci. Technol. 2, 415-418. Hobbs, SE and WW Wolf (1989). An airborne radar technique for studying insect migration. Bulletin of Entomological Research 79(4), 693-704. Hobbs, SE and WW Wolf (1996). Developments in airborne entomological radar. Journal of Atmospheric and Oceanic Technology 13, 58-61. Hobbs, SE and AC Aldhous (2006). Insect ventral radar cross-section polarisation dependence measurements for radar entomology. IEE Proceedings - Radar, Sonar and Navigation 153(6), 502-208. Hunter, DM (1981a). Forecasting migrations of the Australian plague locust. In Annual Report Research Supplement 1979-1980, pp. 62-67. Canberra: Australian Plague Locust Commission. Hunter, DM (1981b). Mass take-off after sunset in the Australian plague locust. In Annual Report Research Supplement 1979-1980, pp. 68-72. Canberra: Australian Plague Locust Commission.

228 Bibliography

Hunter, DM (1982a). Adult development in the Australian plague locust, Chortoicetes terminifera (Walker) (Orthoptera: Acrididae). Bulletin of Entomological Research 72, 589- 598. Hunter, DM (1982b). Diapause development in eggs of the Australian plague locust. In Annual Report Research Supplement 1980-1981, pp. 34-40. Canberra: Australian Plague Locust Commission. Hunter, DM (1989). The response of Mitchell grasses (Astrebla spp.) and Button grass (Dactyloctenium radulans (R. Br.)) to rainfall and their importance to the survival of the Australian plague locust, Chortoicetes terminifera (Walker), in the arid zone. Australian Journal of Ecology 14, 467-471. Hunter, DM (1997). Vitellogensis in Austracris guttulosa (Walker) (Orthoptera: Acrididae) in Tropical Australia. Australian Journal of Entomology 36, 307-308. Hunter, DM (2004). Advances in the control of locusts (Orthoptera: Acrididae) in eastern Australia: from crop protection to preventive control. Australian Journal of Entomology 43, 293-303. Hunter, DM and MD Melville (1994). The rapid and long-lasting growth of grasses following small falls of rain on stony downs in the arid interior of Australia. Australian Journal of Ecology 19, 46-51. Hunter, DM and ED Deveson (2002). Forecasting and management of migratory pests in Australia. Entomologia Sinica 9(4), 13-25. Hunter, DM, RJ Milner, JC Scanlan and PA Spurgin (1999). Aerial treatment of the Migratory Locust, Locusta migratoria (L.) (Orthoptera: Acrididae) with Metarhizium anisopliae (Deuteromycotina: Hyphomycetes) in Australia. Crop Protection 18, 699-704. Hyatt, CJ and DW Maughan (1994). Fourier analysis of wing beat signals: assessing the effects of genetic alterations of flight muscle structure in Diptera. Biophysical Journal 67, 1149- 1154. Irwin, ME (1999). Implications of movement in developing and deploying integrated pest management strategies. Agricultural and Forest Meteorology 97(4), 235-248. Irwin, ME and JM Thresh (1988). Long-range aerial dispersal of ceral aphids as virus vectors in North America. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 321(1207), 421-446. Isard, SA and SH Gage (2001). Flow of Life in the Atmosphere: an Airscape Approach to Understanding Invasive Organisms. East Lansing: Michigan State University Press. Jacobsen, E (2001). On Local Interpolation of DFT Outputs. http://www.primenet.com. Visited on 05 Mar 2001. Jeger, MJ (1999). Improved undestanding of dispersal in crop pest and disease management: current status and future directions. Agricultural and Forest Meteorology 97, 331-349. Johnson, CG (1957). The distribution of insects in the air and the empirical relation of density to height. Journal of Animal Ecology 26(2), 479-494. Johnson, CG (1969). Migration and Dispersal of Insects by Flight. Lodon: Methuen. Key, KHL (1945). The general ecological characteristics of the outbreak areas and outbreak years of the Australian plague locust (Chortoicetes terminifera Wlk.). Aust. Coun. Sci. Ind. Res. Bull. 186. Kootsookos, PJ (1997). A Review of the Frequency and Tracking Problems, 38 pp: DSTO CRC for Robust and Adaptive Systems. Bibliography 229

Korhonen, K (2000). The IGRF-Applet Homepage. Last updated on 2000. http://www.geo.fmi.fi/MAGN/igrf/. Visited on 20 Aug 2003. Kyung, JM, N Jones, DW Borst and MA Rankin (2004). Increased juvenile hormone levels after long-duration flight in the grasshopper, Melanoplus sanguinipes. Journal of Insect Physiology 50(6), 531-537. Lehmann, F-O (2002). The constraints of body size on aerodynamics and energetics in flying fruit flies: an integrative view. Zoology 105, 287-295. Li, GB, HX Wang and WX Hu (1964). Route of the seasonal migration of the oriental armyworm moth in the eastern part of China as indicated by a three-year result of releasing and recapturing of marked moths. Acta Phytophylacica Sinica 3(2), 101-109. Liebhold, AM, RE Rossi and WP Kemp (1993). Geostatistics and geographic information systems in applied insect ecology. Annual Review of Entomology 38, 303-327. Lingren, PD and WW Wolf (1982). Nocturnal activity of the tobacco budworm and other insects. In Biometeorology in Integrated Pest Management, JL Hatfield and IS Thomanson (eds), pp. 211-228. New York: Academic Press. Lingren, PD, JR Raulston, TJ Henneberry and AN Sparks (1986). Night-vision equipment, reproductive biology, and nocturnal behavior: importance to studies of insect flight, dispersal, and migration. In Insect Flight: Dispersal and Migration, W Danthanarayana (ed), pp. 253-264. Berlin: Springer-Verlag. Lockhart, RA and MA Stephens (1985). Tests of fit for the von Mises distribution. Biometrika 72(3), 647-652. Loper, GM and WW Wolf (1986). Using the USDA-ARS insect radar unit to monitor honey bee drone activity. American Bee Journal 126(12), 832. Loper, GM, WW Wolf and ORJ Taylor (1987). Detection and monitoring of honeybee drone congregation areas by radar. Apidologie 18(2), 163-172. Loper, GM, WW Wolf and ORJ Taylor (1988). 1988 Update on the use of a ground-based X- band radar to study the flight of honey-bee drones. American Bee Journal 128(12), 805. Loper, GM, WW Wolf and ORJ Taylor (1989). Update on the use of a ground-based X-band radar to study the flight of honey-bee drones. American Bee Journal 128(12), 805. Loper, GM, WW Wolf and ORJ Taylor (1990). Summary of studies attempting to influence honey bee drone flight direction - radar observations. American Bee Journal 130, 804. Loper, GM, WW Wolf and ORJ Taylor (1993). Radar detection of drones responding to honeybee queen pheromone. Journal of Chemical Ecology 19(9), 1929-1938. Loxdale, HD and G Lushai (2001). Use of genetic diversity in movement studies of flying insects. In Insect Movement: Mechanisms and Consequences, I Woiwood, DR Reynolds and CD Thomas (eds), pp. 361-386. Oxon: CABI Publishing. Lu, SL, CE Lin and CL Huang (1998). Power frequency harmonic measurement using integer periodic extension method. Electric Power Systems Research 44, 107-115. Ma, S-C (1958). The population dynamics of the Oriental Migratory Locust (Locusta migratoria manilensis Meyen) in China. Acta Entomologica Sinica 8(1), 1-40. Magor, JI (1995). Forecasting migrant insect pests. In Insect Migration: Tracking Resources through Space and Time, VA Drake and AG Gatehouse (eds), pp. 399-426. Cambridge: Cambridge University Press.

230 Bibliography

Magor, JI and J Pender (1997). Desert Locust forecaster's GIS: a researchers' view. In New Strategies in Locust Control, S Krall, R Peveling and D Ba Diallo (eds), pp. 21-28. Basel, Switzerland: Birkhäuser Verlag. Mardia, KV (1972). Statistics of Directional Data. London and New York: Academic Press. Mardia, KV and PE Jupp (2000). Directional Statistics. Chichester, New York, Weinheim, Brisbane, Singapore, Toronto: John Wiley & Sons, Ltd. Mazumdar, S, NS Bhaskara Rao and GR Gupta (1965). Radar and Synoptic Study of Locust Swarms over Delhi. Technical Note 69, 162-186 pp: World Met. Org. McCracken, GF and JK Westbrook (2002). Bat patrol. National Geographic(4), 114-123. McDonald, G (1995). Insect migration in an arid continent. I. The common armyworm Mythimna convecta in eastern Australia. In Insect Migration: Tracking Resources through Space and Time, VA Drake and AG Gatehouse (eds), pp. 131-150. Cambridge: Cambridge University Press. McDonald, G, KP Bryceson and RA Farrow (1990). The development of the 1983 outbreak of the common armyworm, Mythimna convecta, in eastern Australia. Journal of Applied Ecology 27(3), 1001-1019. Moskowitz, D, J Moskowitz, S Moskowitz and H Moskowitz (2001). Notes on a large dragonfly and butterfly migration in New Jersey. Northeastern Naturalist 8(4), 483-490. Mueller, EA and EJ Silha (1978). Unique features of the CHILL radar system. In 18th Conference on Radar Meteorology, Atlanta, Ga., Mar 28-31, 1978, pp. 381-382. Boston, Mass.: American Meteorological Society. Mueller, EA and RP Larkin (1985). Insects observed using dual-polarization radar. Journal of Atmospheric and Ocienic Technology 2(1), 49-54. Nash, JC (1979). Compact Numerical methods for Computers: Linear Algebra and Function Minimisation. Bristol: Adam Hilgar Ltd. Natural Resources Canada (2008). Long term movement of the north magnetic pole. Last updated on 2008-01-16. http://gsc.nrcan.gc.ca/geomag/nmp/long_mvt_nmp_e.php. Visited on 09 Oct 2008. Nieminen, M, M Leskinen and J Helenius (2000). Doppler radar detection of exceptional mass- migration of aphids into Finland. International Journal of Biometeorology 44, 172-181. Odiyo, PO (1990). Progress and developments in forecasting outbreaks of the African armyworm, a migrant moth. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 328(1251), 555-569. Osborne, JL, NL Carreck, JR Riley, AD Smith, DR Reynolds and AS Edwards (1997). Harmonic radar: a new technique for investigating bumble bee and honey bee foraging flight. Acta Horticulture 937, 159-163. Osborne, JL, SJ Clark, RJ Morris, IH Williams, JR Riley, AD Smith, DR Reynolds and AS Edwards (1999). A landscape-scale study of bumble bee foraging range and constancy, using harmonic radar. Journal of Applied Ecology 36, 519-533. Pair, SD, JR Raulston, DR Rummel, JK Westbrook, WW Wolf, AN Sparks and MF Schuster (1987). Development and production of corn earworm and fall armyworm in the Texas High Plains: evidence for reverse fall migration. Southwestern Entomologist 12(2), 89-99. Pedgley, DE (1990). Concentration of flying insects by the wind. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 328(1251), 631-653. Bibliography 231

Pedgley, DE, DR Reynolds, JR Riley and MR Tucker (1982). Flying insects reveal small-scale wind. Weather 37(10), 295-306. Pener, MP and Y Yerushalmi (1998). The physiology of locust phase polymorphism: an update. Journal of Insect Physiology 44(5-6), 365-377. Pener, MP, A Ayali and E Golenser (1997). Adipokinetic hormone and flight fuel related characteristics of density-dependent locust phase polymorphism: A review. Comparative Biochemistry and Physiology Part B: Biochemistry and Molecular Biology 117(4), 513-524. Perez, SM, OR Taylor and R Jander (1997). A Sun compass in monarch butterflies. Nature 387, 29. Perez, SM, OR Taylor and R Jander (1999). The effect of a strong magnetic field on Monarch butterfly (Danaus plexippus) migratory behavior. Naturwissenschaften 86, 140-143. Press, WH, SA Teukolsky, WT Vetterling and BP Flannery (1998). Numerical Recipes in C: the Art of Scientific Computing. Cambridge: Cambridge University Press. Pruess, KP and NC Pruess (1971). Telescopic observation of the Moon as a means for observing migration of the army cutworm, Chorizagrotis auxiliaris (Lipidoptera: Noctuidae). Ecology 52(6), 999-1007. Puri, K, GS Dietachmayer, GA Mills, NE Davidson, RA Bowen and LW Logan (1998). The new BMRC limited area prediction system, LAPS. Australian Meteorological Magazine 47(3), 203-223. Quinn, BG (1994). Estimting frequency by interpolation using Fourier coefficients. IEEE Transactions on Signal Processing 42(5), 1264-1268. Rainey, RC (1955). Observation of Desert Locust swarms by radar. Nature 175(4445), 77. Rainey, RC (1989). Migration and Meteorology: Flight Behaviour and the Atmospheric Environment of Locusts and Other Migrant Pests. New York: Oxford University Press. Ramana Murty, BV, AK Roy, KR Biswas and LT Khemani (1964). Observations on flying locusts by radar. J. Sci. Industr. Res. 23, 289-296. Rankin, MA and JCA Burchsted (1992). The cost of migration in insects. Annual Review of Entomology 37, 533-559. Reid, DG, KG Wardhaugh and J Roffey (1979). Radar Studies of Insect Flight at Benalla, Victoria, in February 1974. Technical Paper 16, 1-21 pp. Canberra: CSIRO: Division of Entomology. Reynolds, DR (1988). Twenty years of radar entomology. Antenna 12(2), 44-49. Reynolds, DR (1998). Radar techniques and insect pest management: Why use radar? Tropical Agriculture Association Newsletter 18(4), 31-32 & 7. Reynolds, DR and JR Riley (1979). Radar observations of concnetrations of insects above a river in Mali, West Africa. Ecological Entomology 4, 161-174. Reynolds, DR and JR Riley (1988). A migration of grasshoppers, particularly Diabolocatantops axillaris (Thunberg) (Orthoptera; Acrididae), in the West African Sahel. Bulletin of Entomological Research 78(2), 251-271. Reynolds, DR and JR Riley (1997). Flight Behaviour and Migration of Insect Pests: Radar Studies in Developing Countries. Bulletin 71, 114 pp. Chatham, UK: Natural Resources Institute: Natural Resources Institute. Reynolds, DR, AD Smith and JW Chapman (2008). A radar study of emigratory flight and layer formation by insects at dawn over southern Britain. Bulletin of Entomological Research 98, 35-52.

232 Bibliography

Reynolds, DR, JW Chapman, AS Edwards, AD Smith, CR Wood, JF Barlow and IP Woiwod (2005). Radar studies of the vertical distribution of insects migrating over southern Britain: the influence of temperature inversions on nocturnal layer concentrations. Bulletin of Entomological Research 95, 259-274. Richter, JH, DR Jensen, VR Noonkester, JB Kreasky, MW Stimmann and WW Wolf (1973). Remote radar sensing: atmospheric structure and insects. Science 180, 1176-1178. Riley, JR (1973). Angular and temporal variations in the radar cross-sections of insects. Proc Inst Electr Eng (Lond) 120(10), 1229-1232. Riley, JR (1974). Radar observations of individual Desert Locusts (Schistocerca gregaria (Forsk.) (Orthoptera, Locustidae)). Bull. ent. Res. 64, 19-32. Riley, JR (1975). Collective orientation in night-flying insects. Nature 253(5487), 113-114. Riley, JR (1978). Quantitative analysis of radar returns from insects. In Radar, Insect Population Ecology, and Pest Management, Wallops Flight Center, Wallops Island, Va, May 2-4, 1978, CR Vaughn, WW Wolf and W Klassen (eds), pp. 131-158. NASA. Riley, JR (1979). Radar as an aid to the study of insect flight. In A Handbook on Biotelemetry and Radio Tracking, CJJ Amlaner and DW MacDonald (eds), pp. 131-140. Oxford: Pergamon Press. Riley, JR (1985). Radar cross section of insects. Proc Inst Electr Eng (Lond) 73(2), 228-232. Riley, JR (1988). Recent developments in the application of radar to biology: millimetric and 3- cm systems for studies of insect flight. In Biotelemetry X: Proceedings of the Tenth International Symposium on Biotelemetry, Fayetteville, Arkansas, July 31 - August 15, 1988, CJJ Amlaner (ed), pp. 210-214. Fayetteville: University of Arkansas Press. Riley, JR (1989). Orientation by high-flying insects at night: observations and theories. In Orientation and Navigation - Birds, Humans and Other Animals at 1989 Royal Institute of Navigation, Cardiff, 6-8 April 1989, pp. 1-8. London: Royal Institute of Navigation. Riley, JR (1990). Radar and satellite imagery in insect pest control. Chemistry and Industry 8, 254-257. Riley, JR (1992). A millimetric radar to study the flight of small insects. Electronics & Communication Engineering Journal 4(1), 43-48. Riley, JR (1994). Acoustic sounding, atmospheric structure and insects. International Journal of Remote Sensing 15(2), 293-297. Riley, JR (1999). Riding on the Wind: a Radar Perspective of Insect Flight. London: University of Greenwich. Riley, JR and DR Reynolds (1979). Radar-based studies of the migratory flight of grasshoppers in the middle Niger area of Mali. In Proceedings of the Royal Society of London - Biological Sciences, London, 29-30 June 1977, 204, pp. 67-82. The Royal Society. Riley, JR and DR Reynolds (1983). A long-range migration of grasshoppers observed in the Sahelian zone of Mali by two radars. Journal of Animal Ecology 52, 167-183. Riley, JR and DR Reynolds (1986). Orientation at night by high-flying insects. In Insect Flight: Dispersal and Migration, W Danthanarayana (ed), pp. 71-87. Berlin: Springer-Verlag. Riley, JR and DR Reynolds (1990). Nocturnal grasshopper migration in West Africa: transport and concentration by the wind, and the implications for air-to-air control. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 328(1251), 655- 672. Bibliography 233

Riley, JR and DR Reynolds (1996). Preferential flight by Desert Locusts on favourable winds. In 22nd conference on agricultural and forest meteorology with symposium on fire and forest meteorology and 12th Conference on Biometeorology and Aerobiology, Atlanta, Georgia, January 28 - February 2, 1996, pp. 436-437. Boston: American Meteorological Society. Riley, JR and AS Edwards (1997). Airborne insects as potential sources of acoustic reflectivity. Journal of Atmospheric and Oceanic Technology 14, 538-542. Riley, JR and DR Reynolds (1997). Vertical-looking radar as a means to improve forecasting and control of Desert Locusts. In New Strategies in Locust Control, S Krall, R Peveling and D Ba Diallo (eds), pp. 47-54. Basel, Switzerland: Birkhäuser Verlag. Riley, JR and AD Smith (1999). The Feasibility of Using Vertical-Looking Radar to Monitor the Migration of Brown Planthopper and other Insect Pests of Rice, 1-38 pp. Malvern, Worcestershire: Radar Unit of the Natural Resources Institute (University of Greenwich). Riley, JR, DR Reynolds and MJ Farmery (1981). Radar Observations of Spodoptera exempta Kenya, March - April 1979. Miscellaneous Report 54, 43 pp. London: Centre for Overseas Pest Research. Riley, JR, DR Reynolds and MJ Farmery (1983). Observations of the flight behaviour of the armyworm moth, Spodoptera exempta, at an emergence site using radar and infra-red techniques. Ecological Entomology 8, 395-418. Riley, JR, DR Reynolds and RA Farrow (1987). The migration of Nilaparvata lugens (Stal) (Delphacidae) and other Hemiptera associated with rice during the dry season in the Philippines: a study using radar, visual observations, aerial netting and ground trapping. Bull. ent. Res. 77, 145-169. Riley, JR, AD Smith and BW Bettany (1990a). The use of video equipment to record in three dimensions the flight trajectories of Heliothis armigera and other moths at night. Physiological Entomology 15, 73-80. Riley, JR, AD Smith and RD Gregory (1992a). A radar technique for automatically monitoring insect migration and bio-diversity. In Twelfth International Symposium on Biotelemetry, Ancona, Italy, August 31 - September 5, 1992, M Paolo, F Sandro, C Cristina and B Remo (eds), pp. 310-317. Riley, JR, DR Reynolds and AD Smith (2000). Seasonal monitoring of the insect fauna flying at altitude over southern England: results from a vertical-looking radar. In XXI international Congress of Entomology, Foz do Iguassu, Brazil, 20-26 August 2000, DL Gazzoni (ed). Londrina: Embrapa Soja. Riley, JR, AD Smith and DR Reynolds (2003). The feasibility of using vertical-looking radar to monitor the migration of brown planthopper and other insect pests of rice in China. Entomologia Sinica 10(1), 1-19. Riley, JR, U Krueger, CM Addison and M Gewecke (1988). Visual detection of wind-drift by high-flying insects at night: a laboratory study. Journal of Comparative Physiology, A Sensory, Neural, and Behavioral Physiology 162(6), 793-798. Riley, JR, NJ Armes, DR Reynolds and AD Smith (1992b). Nocturnal observations on the emergence and flight behaviour of Helicoverpa armigera (Lepidoptera: Noctuidae) in the post-rainy season in central India. Bulletin of Entomological Research 82(2), 243-256. Riley, JR, DR Reynolds, S Mukhopadhyay, MR Ghosh and TK Sarkar (1995a). Long-distance migration of aphids and other small insects in northeast India. European Journal of Entomology 92(4), 639-653.

234 Bibliography

Riley, JR, J Holt, DR Reynolds, BT Grayson, MB Green and LG Copping (1990b). The role of migration in the pest status of brown planthopper in temperate and tropical areas. Pest management in rice, 166. Riley, JR, P Valeur, AD Smith, DR Reynolds, GM Poppy and C Lofstedt (1998). Harmonic radar as a means of tracking the pheromone-finding and pheromone-following flight of male moths. Journal of Insect Behavior 11(2), 287-296. Riley, JR, DR Reynolds, AD Smith, AS Edwards, JL Osborne, IH Williams and HA McCartney (1999). Compensation for wind drift by bumble-bees. Nature 400, 126. Riley, JR, AD Smith, DR Reynolds, AS Edwards, JL Osborne, IH Williams, NL Carreck and GM Poppy (1996). Tracking bees with harmonic radar. Nature 379, 29-30. Riley, JR, DR Reynolds, AD Smith, XN Cheng, XX Zhang, GM Xu, JY Cheng, AD Bao and BP Zhai (1990c). Using radar to observe brown planthopper (BPH) migration in China. International Rice Research Newsletter 15(2), 29-30. Riley, JR, XN Cheng, XX Zhang, DR Reynolds, GM Xu, AD Smith, JY Cheng, AD Bao and BP Zhai (1991). The long-distance migration of Nilaparvata lugens (Stal) (Delphacidae) in China: radar observations of mass return flight in the autumn. Ecological Entomology 16(4), 471-489. Riley, JR, DR Reynolds, AD Smith, AS Edwards, XX Zhang, XN Cheng, HK Wang, JY Cheng and BP Zhai (1995b). Observations of the autumn migration of the rice leaf roller Cnaphalocrocis medinalis (Lepidoptera: Pyralidae) and other moths in eastern China. Bulletin of Entomological Research 85(3), 397-414. Riley, JR, DR Reynolds, AD Smith, LJ Rosenberg, XN Cheng, XX Zhang, GM Xu, JY Cheng, AD Bao, BP Zhai and HK Wang (1994). Observations on the autumn migration of Nilaparvata lugens (Homoptera: Delphacidae) and other pests in east central China. Bulletin of Entomological Research 84(3), 389-402. Rinehart, RE (1997). Radar for meteorologiests. Grand Forks, ND: Rinehart Publications. Rizzuto, D (1997). An Easy Way to Call Java Applets with JavaScript. http://www.netscapeworld.com/netscapeworld/nw-05-1997/nw-05-jjscript.html. Visited on 24/05/2000. Robinson, TP (1995). Geographic information systems and remotely sensed data for determining the seasonal distribution of habitats of migrant insect pests. In Insect Migration: Tracking Resources through Space and Time, VA Drake and AG Gatehouse (eds), pp. 335-352. Cambridge: Cambridge University Press. Rochester, WA (1999). The Migration Systems of Helicoverpa punctigera (Wallengren) and Helicoverpa armigera (Hübner) (Lepidoptera: Noctuidae) in Australia. PhD Thesis. Department of Zoology and Entomology, University of Queensland. Rochester, WA and MP Zalucki (1998). Measuring the impacts of Helicoverpa armigera migration on pest management during summer and autumn. In Pest Management - Future Challenges. Proceedings of the Sixth Australasian Applied Entomological Research Conference, Brisbane, Australia, 28 September - 2 October 1998, MP Zalucki, RAI Drew and GG White (eds), 2, pp. 94-98. The University of Queensland Printery. Rochester, WA, ML Dillon, GP Fitt and MP Zalucki (1996). A simulation model of the long- distance migration of Helicoverpa spp. moths. Ecological Modelling 86(2-3), 151-156. Roffey, J (1969). Radar Studies on the Desert Locust, Niger, September - October 1968. Occasional Report 17, 1-14 pp. London: Anti-locust Research Centre. Roffey, J (1972). Radar studies of insects. Pest Abstr. and News Summ. 18(3), 303-309. Bibliography 235

Rose, DJW, WW Page, CF Dewhurst, JR Riley, DR Reynolds, DE Pedgley and MR Tucker (1985). Downwind migration of the African armyworm moth, Spodoptera exempta, studied by mark-and-capture and by radar. Ecological Entomology 10, 299-313. Rosenberg, LJ and JI Magor (1987). Predicting windborne displacements of brown planthopper, Nilaparvata lugens from synoptic weather data. Journal of Animal Ecology 56, 39-51. Russell, RW and JW Wilson (1997). Radar-observed "fine lines" in the optically clear boundary layer: reflectivity conributions from aerial plankton and its predators. Boundary Layer Meteorology 82, 235-262. Schaefer, GW (1967). Recent studies of bird flight by radar. British Ornithologists' Union 109, 476. Schaefer, GW (1969). Radar studies of locust, moth and butterfly migration in the Sahara. In Royal Entomological Society of London, Lodon, 3rd Dec 1969, 34, pp. 33, 39-40. Schaefer, GW (1970). Radar detection of individual locusts and swarms. In Int. Study Conf. Current and Future Problems of Acridology, London, pp. 379-380. Centre for Overseas Pest Research. Schaefer, GW (1976). Radar observations of insect flight. In Insect flight, RC Rainey (ed), pp. 157-197. Oxford: Blackwell Scientific. Schaefer, GW (1979). An airborne radar technique for the investigation and control of migrating insect pests. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 287(1022), 459-465. Schaefer, GW, GA Bent and R Cannon (1979). The green invasion. New Scientist 83, 440-441. Schneider, M, G Wiesel and A Dorn (1995). Effects of JH III and JH analogues on phase- related growth, egg maturation and lipid metabolism in Schistocerca gregaria females. Journal of Insect Physiology 41(1), 23-31. Schregardus, DS, AW Pieneman, AT Maat, RF Jansen, TJF Brouwer and ML Gahr (2006). A lightweight telemetry system for recording neuronal activity in freely behaving small animals. Journal of Neuroscience Methods 155(1), 62-71. Sedlacek, M and M Titera (1998). Interpolations in frequency and time domains used in FFT spectrum analysis. Measurement 23(3), 185-193. Sherlock, BG (1999). Windowed discrete Fourier transform for shifting data. Signal Processing 74(2), 169-177. Shields, EJ and AM Testa (1999). Fall migratory flight initiation of the potato leafhopper, Empoasca fabae (Homoptera: Cicadellidae): observations in the lower atmosphere using remote piloted vehicles. Agricultural and Forest Meteorology 97(4), 317-330. Sincich, T (1993). Statistics by Example. New York: Dellen. Smith, AD and JR Riley (1996). Signal processing in a novel radar system for monitoring insect migration. Computers and Electronics in Agriculture 15(4), 267-278. Smith, AD, JR Riley and RD Gregory (1993). A method for routine monitoring of the aerial migration of insects by using a vertical-looking radar. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 340(1294), 393-404. Smith, AD, DR Reynolds and JR Riley (2000). The use of vertical-looking radar to continuously monitor the insect fauna flying at altitude over southern England. Bulletin of Entomological Research 90(3), 265-277. Smith, WW and JM Smith (1995). Handbook of real-time fast Fourier transforms: algorithms to product testing. New York: The Institute of Electrical and Electronics Engineers, Inc.

236 Bibliography

Sparks, AN, SD Pair, JK Westbrook and WW Wolf (1987). Migration team tracks night flyers. Agricultural Research 35(7), 6-9. Sparks, AN, JK Westbrook, WW Wolf, SD Pair and JR Raulston (1985). Atmospheric transport of biotic agents on a local scale, 203-216. Spyers-Ashby, JM, PG Bain and SJ Roberts (1998). A comparison of fast fourier transform (FFT) and autoregressive (AR) spectral estimation techniques for the analysis of tremor data. Journal of Neuroscience Methods 83(1), 35-43. Srinivasan, MV, M Poteser and K Kral (1999). Minireview: Motion detection in insect orientation and navigation. Vision Research 39(2749-2766). Srygley, RB (2003). Wind drift compensation in migrating dragonflies Pantala (Odonata: Libellulidae). Journal of Insect Behavior 16(2), 217-232. Srygley, RB and EG Oliveira (2001). Orientation mechanisms and migration strategies within the flight boundary layer. In Insect Movement: Mechanisms and Consequences, I Woiwood, DR Reynolds and CD Thomas (eds), pp. 183-206. Oxon: CABI Publishing. Srygley, RB, EG Oliveira and R Dudley (1996). Wind drift compensation, flyways, and conservation of diurnal, migrant Neotropical Lepidoptera. In Proceedings of the Royal Society of London - Biological Sciences, London, 263, pp. 1351-1357. The Royal Society. Stefanescu, C, M Alarcón and A Àvila (2007). Migration of the painted lady butterfly, Vanessa cardui, to north-eastern Spain is aided by African wind currents. Journal of Animal Ecology 76(5), 888-898. Stephens, MA (1965). The goodness-of-fit statistic Vn: distribution and significance points. Biometrika 52, 309-321. Svensson, GP, PG Valeur, DR Reynolds, AD Smith, JR Riley, TC Baker, GM Poppy and C Lofstedt (2001). Mating disruption in Agrotis segetum monitored by harmonic radar. Entomologia Experimentalis et Applicata 101(2), 111-121. Tapper, N and L Hurry (1993). Australian Weather Patterns: An Introductory Guide. Mount Waverley, Victoria: Dellasta Pty Ltd. Taylor, LR (1974). Insect migration, flight periodicity and the boundary layer. Journal of Animal Ecology 43(1), 225-238. Tucker, MR (1997). Satellite-derived rainstorm distribution as an aid to forecasting African armyworm outbreaks. Weather 52(7), 204-212. Upton, GJG and B Fingleton (1985). Spatial Data Analysis by Example. Vol 2: Categorical and Directional Data. Chichester: John Wiley & Sons. Voss, F and U Dreiser (1997). Mapping of Desert Locust habitats using remote sensing techniques. In New Strategies in Locust Control, S Krall, R Peveling and D Ba Diallo (eds), pp. 37-45. Basel, Switzerland: Birkhäuser Verlag. Walker, MM, TE Dennis and JL Kirschvink (2002). The magnetic sense and its use in long- distance navigation by animals. Current Opinion in Neurobiology 12(6), 735-744. Wang, HK and VA Drake (2004). Insect monitoring radar: retrieval of wingbeat information from conical-scan observation data. Computers and Electronics in Agriculture 43(3), 209- 222. Wang, HK, BH Zhou and XN Cheng (1997). Geographic information system and its application on the pest management. Entomological Knowledge 34(6), 366-370. Westbrook, JK and WW Wolf (1998). Migratory flights of bollworms, Helicoverpa zea (Boddie), indicated by doppler weather radar. In Second Urban Environment Symposium & Bibliography 237

13th Conference on Biometeorology and Aerobiology, Albuquerque, New Mexico, 2-6 November 1998, pp. 354-355. Westbrook, JK and SA Isard (1999). Atmospheric scales of biotic dispersal. Agricultural and Forest Meteorology 97(4), 263-274. Westbrook, JK, RS Eyster, WW Wolf, PD Lingren and JR Raulston (1995a). Migration pathways of corn earworm (Lepidoptera: Noctuidae) indicated by tetroon trajectories. Agricultural and Forest Meteorology 73(1-2), 67-87. Westbrook, JK, JR Raulston, WW Wolf, SD Pair, RS Eyster and PD Lingren (1995b). Field observations and simulations of atmospheric transport of noctuids from northeastern Mexico and the south-central U.S. Southwestern Entomologist(Supp 18), 25-44. Wikelski, M, D Moskowitz, JS Adelman, J Cochran, DS Wilcove and ML May (2006). Simple rules guide dragonfly migration. Biology Letters 2(3), 325-329. Wikelski, M, RW Kays, NJ Kasdin, K Thorup, JA Smith and GW Jr Swenson (2007). Going wild: what a global small-animal tracking system could do for experimental biologists. The Journal of Experimental Biology 210, 181-186. Wilkie, D (1983). Rayleigh test for randomness of circular data. Applied Statistics 32(3), 311- 312. Williams, IH, JL Osborne, JR Riley, AD Smith, DR Reynolds, NL Carreck and GM Poppy (1999). Harmonic radar: progress in using this new technique to study bee flight. In 6th International Conference on Apiculture in Tropical Climates, Costa Rica, 18-19 August 1999, pp. 135-139. Cardiff: International Bee Research Association. Wolda, H (1988). Insect seasonality: why? Annual Review of Ecology and Systematics 19, 1-18. Wolf, WW (1979a). Entomological radar studies in the United States [Insect migration]. In Movement of highly mobile insects: concepts and methodology in research; proceedings of Movement of Selected Species of Lepidoptera in the Southeastern United States, RL Rabb and GG Kennedy (eds), pp. 263-266. Raleigh, North Carolina: University Graphics, North Carolina State University. Wolf, WW (1979b). Radar program at Western Cotton Research Laboratory [U.S. Department of Agriculture, for studying insect dispersal]. In Radar, Insect Population Ecology, and Pest Management, CR Vaughn, WW Wolf and W Klassen (eds), pp. 159-160. Wallops Island, Va: Wallops Flight Center, NASA. Wolf, WW and SD Pair (1982). Entomological radar: observations during 1981. In Summer Meeting of American Society of Agricultural Engineers, University of Wisconsin-Madison, June 27-30, 1982, pp. 1-21. Wolf, WW and PG Schleider (2002). Entomological Radar Studies at USDA ARS APMRU. http://scrl.usda.gov/SCRL/apmru/imms/radar/index.htm. Visited on 6 Jun 2002. Wolf, WW, JK Westbrook and AN Sparks (1986a). Relationship between radar entomological measurements and atmospheric structure in south Texas during March and April 1982. In Long-range migration of moths of agronomic importance to the United States and Canada: Specific examples of occurrence and synoptic weather patterns conductive to migration., AN Sparks (ed), pp. 84-97: USDA, ARS. Wolf, WW, GM Loper and G Greneker (1987). Doppler radar for detecting insects. In Summer Meeting of American Society of Agricultural Engineers, Baltimore Convention Center, Baltimore, MD, June 28-July 1, 1987, pp. 1-19. American Society of Agricultural Engineers. Wolf, WW, CR Vaughn, R Harris and GM Loper (1993a). Insect radar cross-sections for aerial density measurements and target classification. Transactions of the ASAE 36(3), 949-954.

238 Bibliography

Wolf, WW, JK Westbrook, JR Raulston, SD Pair and SE Hobbs (1990). Recent airborne radar observations of migrant pests in the United States. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 328(1251), 619-630. Wolf, WW, JK Westbrook, JR Raulston, SD Pair and PD Lingren (1993b). Radar detection of ascent of Helicoperpa zea (Lepidoptera: Noctuidae) moths from corn in the lower Rio Grande valeey of Texas. In 13th Internation Congress of Biometeorology, Calgary, Alberta, Canada, 12-18 September 1993, pp. 1065-1074. Wolf, WW, AN Sparks, SD Pair, JK Westbrook, FM Truesdale and W Danthanarayana (1986b). Radar observations and collections of insects in the Gulf of Mexico. In Insect Flight: Dispersal and Migration, pp. 221-234. Berlin: Springer-Verlag. Wood, CR, JW Chapman, DR Reynolds, JF Barlow, AD Smith and IP Woiwod (2006). The influence of the atmospheric boundary layer on nocturnal layers of noctuids and other moths migrating over southern Britain. International Journal of Biometeorology 50, 193-204. Wootton, RJ (2001). How insect wings evolved. In Insect Movement: Mechanisms and Consequences, I Woiwood, DR Reynolds and CD Thomas (eds), pp. 43-64. Oxon: CABI Publishing. Wright, DE (1987). Analysis of the development of major plagues of the Australian plague locust Chortoicetes terminifera (Walker) using a simulation model. Australian Journal of Ecology 12, 423-438. Wright, DE and PM Symmons (1987). The development and control of the 1984 plague of the Australian plague locust, Chortoicetes terminifera (Walker). Crop Protection 6(1), 13-19. Yang, XL, L Chen, DF Cheng and JR Sun (2008). Primary application of millimetric scanning radar to tracking high-flying insects in southern China. Plant Protection 34(2), 31-36. Zalucki, MP, DAH Murray, PC Gregg, GP Fitt and PH Twine (1994). Ecology of Helicoverpa armigera (Hubner) and H. punctigera (Wallengren) in the inland of Australia: larval sampling and host plant relationships during winter and spring. Aust. J. Zool. 42, 329-346. Zera, AJ and RF Denno (1997). Physiology and ecology of dispersal polymorphism in insects. Annual Review of Entomology 42, 207-230. Zhai, BP (1999). Tracking angels: 30 years of radar entomology. Acta Entomologica Sinica 42(3), 315-326. Zhai, BP and XX Zhang (1993). Behaviour of migrating insects: adaptation and selection to atmospheric environment. Acta Ecologica Sinica 13(4), 356-363. Zhang, FS, ZX Geng and W Yuan (2001). The algorithm of interpolating windowed FFT for harmonic analysis of electric power system. IEEE Transactions on Power Delivery 16(2), 160-164. Zhang, YH, HB Qiao, DF Cheng and JR Sun (2007a). Primary application of vertical-looking radar to tracking high-flying insects in China. Plant Protection 33(3), 23-26. Zhang, YH, L Chen, DF Cheng, YJ Zhang, YY Jiang and JW Jiang (2007b). Radar observation and population analysis on the migration of the clover cutworm, Scotogramma trifolii Rottemberg (Lepidoptera: Pyralidae). Acta Entomologica Sinica 50(5), 494-500. Zhang, YH, DF Cheng, B Jiang, L Chen, HB Qiao, XL Young and JR Sun (2006). Vertical- looking entomological radar and its aplication. In Science and Technology Innovation and Protection of the Unpolluted Plant, ZM Cheng (ed), pp. 642-646. Beijing, China: Agricultural Scientech & Technology Press. Bibliography 239

Zhang, YH, L Chen, DF Cheng, Z Tian, JR Sun, JY Ying and YJ Zhang (2008). Nocturnal migration of Coleoptera: Carabidae in North China. Agricultural Sciences in China 7(8), 977-986. Zhang, Z-B and D-M Li (1999). A possible relationship between outbreaks of the oriental Migratory Locust (Locusta migratoria manilensis Mayer) in China and the El Nino episodes. Ecological Research 14, 267-270. Zheng, ZQ and XX Zhang (2000). Flight capacity and facultative migration of cotton bollworm. Chinese Journal of Applied Ecology 11(4), 603-608. Zhou, BH, HK Wang and XN Cheng (1995). Forecasting systems for migrant pests. I. The brown planthopper Nilaparvata lugens in China. In Insect Migration: Tracking Resources through Space and Time, VA Drake and AG Gatehouse (eds), pp. 353-364. Cambridge: Cambridge University Press.

9 Appendices

A. Interaction between Java and JavaScript

Equation Chapter 1 Section 1It is necessary to verify the availability of the required data on the web server when generating a concrete web page for a specific date on-the- fly from the abstract Java-JavaScript web template. HTML (Hyper Text Mark-up Language) does not support file checking, and for security reasons JavaScript only supports text/html file checking on the web server. Java provides a means of accomplishing this task. However, Java cannot access either HTML fields or JavaScript variables directly, while JavaScript can only access the public methods (but not the variables) in a Java applet. Netscape Communications Corporation (Mountain View, CA, USA, http://www.netscape.com) implemented the LiveConnect technology to let Java and JavaScript communicate to each other (Rizzuto 1997; Flanagan 1998). Unfortunately, it is not fully available on all browsers and platforms. Gagnon (1999) used a hidden frame to act as a bridge between Java applets and JavaScript functions in different frames, so avoiding use of LiveConnect. This is an ingenious way to use the URL (Universal Resource Locator) format as both Java and JavaScript can form and decode a URL, in which parameters can be encoded. As the page’s content is required to change dynamically to show the observations for different dates, with DHTML (Dynamic HTML) technology, the whole document can be recomposed and reloaded. Therefore a simpler solution has been adopted: using the document URL to pass parameters from JavaScript to Java, without frames being used. A Java applet embedded in the front page searches the document URL for a date that may have been encoded there by a JavaScript function in response to the user’s request. The applet then

241 242 Appendices connects to the web server to check if the data is available for the specified date (by searching for the index-bar GIF file for that date in its appropriate directory), or for the previous night if no date is encoded in the URL. The applet stores the data availability information in a public method which can be accessed from a JavaScript function. However, the following technical difficulties must be solved:

a) Readiness of the Java applet. A Java applet has an inactive time when the page including it is being loaded, because it can only be embedded in the body part of an HTML file and it requires time for download and initialisation. Therefore, a JavaScript function needs to wait until the applet finishes loading and initialising before it can call the applet. Putting the applet before the JavaScript function will not help because the applet may take longer than the JavaScript function to load into memory, as embedded objects are transferred in different threads. Setting a specific time for the JavaScript function to wait does not work either as the time depends on the user’s Internet connection and CPU speed. Frequently checking the applet status requires different routines specified for different browsers, since at least Netscape Navigator/Communicator and Microsoft Internet Explorer respond differently to the applet being completely loaded and initialised (Gagnon 1999). However, it has been found, surprisingly, that the onLoad property of the web page indicates the status of both page loading and applet initialisation. Therefore, a JavaScript function can be triggered by the onLoad event to access the applet’s public method that returns the data availability, and then to compose a new web page according to the data availability.

b) DHTML implementation. Although the two major Internet browsers, Navigator/Communicator and Internet Explorer, have both been declared to support DHTML since version 4, there are implementation differences between them, e.g. for support of frames, layers, object properties, JavaScript and VBScript. Therefore, dynamically changing a page’s content needs to be implemented specifically for the user’s particular browser. A routine that identifies the browser type and checks the availability of JavaScript and Java support is required, and the web page can then be composed on-the-fly in the form appropriate for the user’s browser. A. Interaction between Java and JavaScript 243

A standard web server (URL http://www.imr.ph.adfa.edu.au) has been set up with the Apache HTTP server (Version 1.3.26, http://www.apache.org) on the base- laboratory hub PC. This computer can run under either the Microsoft Windows 98 (Second Edition) or the Slackware Linux (Version 7.0, kernel 2.2.13, http://www.slackware.org) operating system. No additional module, e.g. CGI, PHP (Hypertext Preprocessor), or database server, is required. The observation result files from the IMRs are stored on the network server, and are immediately available on the Internet for public view.

The web page has been built with DHTML, incorporating JavaScript and Java technology as described above. There is one web page for each IMR. The web page (~27 KB) is coded in the JavaScript language with a 2.4KB hidden Java applet embedded. It is actually an empty generic page, serving as a template from which concrete web pages are generated with GIF graphs of the IMR observations for the user- specified dates. When the web page is parsed during the loading stage, the user’s browser is checked for the required JavaScript and Java support, and if this is not found, a warning message will be given and page generation will be terminated. Otherwise, an observation summary page, which is the template page filled with available GIF graphs, will be generated by a JavaScript function for the previous night or the user-specified

index.html Load Java Applet page

JavaScript YYJava Load Get date enabled enabled Java applet from URL

Wait for Check data applet ready NNWarning availability for browser

1 Compile new page

JavaScript Functions

Figure A-1 Flow Chart of Web Generation The front page index.html is written with DHTML and contains client-side JavaScript functions and a Java applet to compose a web page on-the-fly for the date specified by the user.

244 Appendices date. When the user enters a date on the web page, a JavaScript function encodes this date into the URL and forces the page to be regenerated. Thus, all that is required to make the observations available for view on the WWW is to store the observation summaries in the appropriate directories on the server. Figure A-1 shows the flow chart of the interaction between the JavaScript functions and the Java applet when generating a page. B. Migration Events of Chortoicetes terminifera at Bourke 245

B. Migration Events of Chortoicetes terminifera at Bourke

The nights of C. terminifera migration identified with echo signatures from the IMR observations at Bourke during the three-year period June 1998 – May 2001 are listed here. A brief summary of their main properties is included. Classification as emigration, overflight, and immigration has been made according to the patterns of the time- series/vertical-profile plots (see Chapter 5). These nights form the primary dataset for the analyses of the following two chapters.

Table A-1 Migration Nights of Chortoicetes terminifera Detected by the IMR at Bourke Month Migration Night Migration Type Dec/98 31 Emigration to SSW. Jan/99 02-15, 17-20, 28-29 Immigration from SE-NE during first 6 h on 02-06, [Data not available 4 h after sunset from NE on 07 and from SSE on (NA) on 16, 21-27] 08; overflight from SE-NE on 09; immigration from NW after wind shift from NW to SE on 10, from NE during a period of rain when the wind shifted from SW to SE on 11, from SW on 12, and from SW during rain on 13, from E on 14 and 15, from SSE on 17, from ENE on 18, 3 h after sunset from S on 19, and from SE on 20, at low altitudes from ESE on 28, and from SE on 29. Feb/99 04-06, 10-23, 25-26, Immigration from SE on 04 and 05, from W on 06 28 (small numbers on 05, 06), from E on 10, from SE (Raining on 24, 27 on 11-14 (small numbers on 13-14), 4 h after sunset NA) from SSW on 15; immigration from SW earlier and overflight from SSE later on 16; emigration to E on 17, to NNE on 18 with overflight detected afterwards; overflight 4 h after sunset from SE on 19; immigration from SE on 20, small number from ESE on 21; overflight during first 3 h and immigration afterwards from SE on 22; emigration to NNW and immigration from SSW on 23; emigration and overflight to NW on 25, 26 and 28. Mar/99 01, 04, 07, 13-16, Emigration and overflight to NW on 01; emigration 24-27 to NE and overflight to NW on 04; emigration to NNW on 07; small immigration from E on 13; emigration to NNE and immigration from S on 14; overflight from SSE on 15; immigration and overflight from SSE on 16; early immigration and

246 Appendices

Month Migration Night Migration Type late overflight from ENE on 24; small overflight from E on 25 and 26; late immigration from SW after wind shift from NNW to SSW on 27. Nov/99 05-06, 29 Small overflight from NW during late night on 05; short and small immigration from WSW on 06; small immigration from E on 29. Dec/99 03, 07 Immigration from SSE 3 h after sunset on 03, small (04-06, 09-14 NA) immigration from ENE on 07. Jan/00 01-02, 06-08, 11-12, Small and short immigration from SSE during late 25 night on 01; emigration to NW early and (26-31 NA) immigration later from SE on 02; small immigration from SSE on 06, from SE on 07, and from SE after wind shifted from SW to SE on 08; small immigration from SE on 11 and 12; small overflight from ENE on 25. Feb/00 06, 10, 14-15, 23-29 Small immigration from SE 4 h after sunset on 06, (01-05, 16-20 NA) small overflight from NE on 10; small overflight and immigration from SE on 14; immigration from SE on 15; small overflight from ENE on 23; emigration to WNW on 24 and 25; overflight SE on 26 and 27; overflight and immigration from SE on 28; immigration from SE on 29. Mar/00 01-02, 08-09, 20, 23, Small emigration to W on 01, to NW on 02 and 08; 28 immigration from E on 09 before rain; emigration (03-07 NA) to NNE on 20; small emigration to NE and immigration from SSW on 23; small emigration to ENE on 28. Apr/00 04, 15 Overflight and immigration from NW on 04; overflight from NW on 15. Nov/00 08, 12, 14 Overflight from NE on 08; immigration from SW- SE and overflight from NE during wind shift on 12; small overflight from NE on 14. Dec/00 06-08, 12, 21-24 Immigration from NW-NE on 06 when light rain was detected at midnight, from SW-SE on 07 and WSW-SW on 08 during wind shift, from NNW on 12; overflight from NW-N on 21; overflight from NNW and immigration from SSE after wind shift on 22; small immigration from NW after rain on 23; immigration from SW on 24. Jan/01 07, 13, 15, 17, 24, Emigration to NE on 07; immigration from S-SW 26-29 on 13, from SSE on 15; overflight from SW on 17, (01-06 NA) from WSW on 24; emigration to ENE and overflight from WSW on 26; overflight mainly B. Migration Events of Chortoicetes terminifera at Bourke 247

Month Migration Night Migration Type from SW before wind shifted through full 360° at low altitude on 27, from SSW-SSE on 28 and 29. Feb/01 01-20, 24-25,27-28 Immigration from SE on 01, ESE on 02; immigration and overflight from ENE on 03; immigration from NE before rain after midnight on 04, from ENE on 05, from SE on 06; overflight from SE on 07; immigration from E on 08, from NE on 09; emigration to NNW, overflight and immigration from SE on 10; early immigration before rain on 11; immigration from E and overflight from NE on 12; emigration to NW and overflight from NE on 13; emigration to NNE and immigration from SSE on 14; emigration to NNE and overflight from SSE on 15; late and short immigration from SE on 16 and 17; emigration to NW on 18; late immigration from SE on 19; emigration to NW and overflight from SE on 20; late and short immigration from N on 24; emigration to NNW on 25; immigration from SW during rain on 27; massive immigration from SSW- SSE on 28. Small numbers on 7-9,12-13, 17-18, 24. Mar/01 01-06, 08-10, 12, 25 Massive immigration from SW-SE during rain on (07 NA) 01; immigration from ENE before wind shear on 02; emigration to W and overflight from E on 03; emigration to NW and immigration from SE on 04, 05 and 06; immigration from SE on 08 and 09; short overflight after midnight on 10; immigration from SSW and overflight from SSE on 12; small immigration from WSW-WNW on 25. Dates are those for the start of the night.

C. Metapopulation Persistence of Plague Locust during 1998-2001 in Eastern Australia 249

C. Metapopulation Persistence of Plague Locust during 1998 – 2001 in Eastern Australia

With IMR observations and trajectories estimated from them, population flows of plague locust within the metapopulation in eastern Australia could be detected, and thus the population development and distribution forecast. In Section 2.2 of Chapter 5, the migrations in the season of 1999 – 2000 were analysed in detail and the sequence of plague population build-up was reconstructed. To provide a full view of the plague development and collapse in eastern Australia, this appendix gives information about this locust outbreak through 1998-2001, covering the whole cycle. From IMR migration evidence, variations of plague locust populations in eastern Australia during the study period are examined, in association with other ancillary information, such as the APLC field-survey, locust control, light-trap, remote-sensing and weather data.

Increase of Population in 1998 – 1999 At the start of the 1998-99 season, generally only a low background locust population was present in eastern Australia. There were almost no bands or swarms in either the arid interior (the Lower Western and Far Southwest districts of Queensland, the Upper Western of New South Wales and the Far North of South Australia), or the agricultural belt of NSW (the North and Central West Plains and Slopes) in autumn 1998. Widespread heavy rainfall in winter ensured high survival of the nymphs from overwintering eggs.

Spring generation: There were low density populations of nymphs in most areas of eastern Australia following the heavy rainfall in early spring. Flooding in the North West Plains of NSW in September may have delayed egg hatching and caused high mortality of nymphs. By late spring, populations were above normal background levels in many regions, from the Riverina of NSW to the Central Lowlands of Queensland. Numerous to sub-band nymphal densities were reported in the Central West Plains and Slopes of NSW, while a low-density swarm was found in the Riverina in mid November. However, little rain fell in the arid interior during the spring.

250 Appendices

Summer generation: Dry conditions limited population increase during the summer. Populations increased slightly but most areas had no bands or swarms. No large scale migrations were detected, though small-scale migrations may have occurred in the Channel Country, populations from the Far Southwest to the Lower Western of Qld in early January, and a swarm near Owangowan in the Warrego to the Far Southwest of Qld in mid-February. Despite some rain fell in central and southwestern Qld, most rain in December was along the coast and tropical Qld. Rain in early January covered the whole of Qld and the coast region of NSW, up to the North and Central West Plains and Slopes. Rainfall in late January covered most of the regions with locust populations except for the Far North of SA. February rainfall was mainly on the coast regions of Qld and NSW, except for the rain in late February to beginning of March, which covered almost the whole of Qld and the northeast of NSW. Again, no rain fell in the arid interior.

Autumn generation: Population levels were above normal in many districts but had not reached levels warranting control. Some small bands were found in the Central and North West Plains and Slopes of NSW and the Warrego of Qld, indicating that the population had increased. These populations apparently migrated southwards in early April into the Central West Plains and Slopes and the Riverina of NSW, where swarms were seen laying overwintering eggs. The next major rain was in late March and early April, when widespread rain fell over much of eastern Australia, created suitable conditions for egg laying. There was no rainfall in the Far North of SA.

During 1998 – 1999, the IMR at Bourke detected migrations of plague locust on about 50 nights during the summer: no migrations were recorded in the spring or autumn periods. Migrations were generally in a westward direction (Figure 5.20). However, the unusually low summer rainfall limited population increase in the arid interior, even though background levels were above normal and populations continuously moved into this area. The late summer and early autumn rains, however, produced a significant overwintering population in the agricultural zone of NSW.

Development of Plague in 1999 – 2000 Good winter rain fell throughout southeastern Australia with the exception of the Far North of SA). The rainfall would have enabled the development of eggs and the survival of nymphs in the Riverina and Central West Plains of NSW in the spring. C. Metapopulation Dynamics of Plague Locust during 1998-2001 in Eastern Australia 251

Spring generation: A major locust population was present in the Riverina and Central West Plains of NSW (Figure A-2a), where the APLC conducted extensive aerial control during October and November spraying 76 band and 107 swarm targets. . However, a significant residual population remained in these areas as not all populations were concentrated in band or swarm “targets” suitable for aerial spraying. No major migration to the interior was detected by the IMR at Bourke, but a small SW-NE movement was observed during 5-7 November. Trajectories at different heights, calculated from the IMR observations for the night of 06-07 November (Figure A-2a) indicate the population in the Riverina might have moved into the Central West Slopes and Plains. In addition, wind trajectories suggested locusts might have moved westward on 17 and 18 November as swarms was seen in many places in Upper Western NSW but not Bourke on 19 November. Further movements to the west and northwest into the interior were possible under the weather conditions in late November, when a small locust migration was detected by the IMR at Bourke on 29 November. The inland population was limited with little rain in winter and early spring; the later APLC field survey confirmed this. Spring rain fell into the Riverina and the Central West Plains ensured the hatching and survival of very large numbers of overwintering eggs. Heavy rainfall in late spring extended over much of eastern Australia, except for the Far North of SA, creating ideal conditions for fat accumulation for migration and breeding.

Summer generation: Field-survey and light-trap data indicated that the adult population increased markedly in the Channel Country of Queensland in late spring- early summer. (Figure A-2b, c). As no locusts were caught by the light-trap at White Cliffs until 30 November, at Fowler Gap until 2 December and at Thargomindah until 4 December, the population possibly migrated in late November to early December from the Central West Plains of NSW, where a significant decrease in numbers was recorded by APLC. The IMR at Bourke detected heavy migrations into the inland on the night of 03-04 December 1999 (Figure A-2b) and numerous locusts and swarm were reported in southwestern Queensland on 5 December. Another migration was detected at Bourke on 02-03 January 2000 (Figure A-2c), which was also indicated by the light-trap catches at Fowler Gap and Thargomindah on 3 January 2000. The APLC carried out aerial spraying of bands and swarms in the Channel Country of Queensland from mid-January, onward with 12 band and 34 swarm targets sprayed by early February. Swarms were also reported in the Far North of SA in mid-January. However, floodwaters from

252 Appendices

Figure A-2 Distributions of Chortoicetes terminifera in 1999 – 2000 Survey data are plotted on NOAA NDVI images as turquoise triangles and squares for nymphs and red circles for adults (APLC data). Trajectory examples (blue – forward, red – backward) are calculated from the IMR observations. C. Metapopulation Dynamics of Plague Locust during 1998-2001 in Eastern Australia 253 the December rain limited field surveys to determine the population levels and age structures. Heavy rainfall and further immigrations allow the populations to increase to very high levels. A population decrease in the Channel Country of Queensland indicated the likely sources of the immigrants. The Far Northwest and Lower Darling of Western NSW also received large number of immigrants, possibly from the Far Southwest of Queensland and/or from the Central West Plains of NSW. Both IMRs at Thargomindah and Bourke detected southwestward and northwestward to westward migrations respectively during late January to early February. The IMRs detected similar movements in late February (trajectories for 24-25 February are shown on Figure A-2d) until early March. The population in the Central West Plains of NSW remained at low levels. The population in the Riverina of NSW was increasing, possibly from hatchings from eggs laid in mid-December by a persisted population. Small bands were reported in the Riverina in late January and 34 swarms were sprayed during late February to early March. Heavy rain fell across much of eastern Australia during December, creating ideal conditions for locust breeding. Continuous rain in January improved conditions for both survival and breeding. Heavy rain fell into the arid interior before mid-February and later into most of Qld, western NSW, Vic and southern SA.

Autumn generation: Heavy rains allowed the population that migrated to the Far Northwest of NSW and the Far North, Northeast and North of SA, to breed successfully leading to a major outbreak in these regions (Figure A-2d, e) in March. A total of 131 band targets were controlled in the Far Northwest of NSW by APLC aerial spraying from 25 March to 3 April. This operation significantly reduced the overall population but many locusts would have remained. In the Far North of SA, however, only 3 bands were sprayed as a result of difficulty in finding bands of suitable size and density for aeriual spraying. Substantial populations were left in this region in conditions that were suitable for breeding. The residual populations in the Riverina of NSW was also still in a favourable environment, and there were also high population levels in adjacent regions. Outbreaks developed in a wide area of the Western NSW, the Far North and Northern Agricultural districts of SA, and the Mallee of Vic. The APLC sprayed 148 swarms in central SA and western NSW, and 23 bands in the Riverina of NSW, between 8 April and 2 May. There were considerable residual populations remaining as control was hampered by adverse weather, and swarm migrations and egg laying were reported. The light-traps at White Cliffs and Fowler Gap recorded frequent migration

254 Appendices activities during April (migration trajectories for the night of 15-16 April are shown on Figure A-2e). These migrations redistributed the populations further south over a large area. Heavy rain fell over most of the locust area, except for the north of SA, in March. Rainfall in mid-April in both the inland and the agricultural regions kept conditions favourable.

The APLC controlled a total of 568 band and swarm targets extensively over 2612 km2, using 54,657 litres of insecticide, during 1999 – 2000. Despite this extensive effort, significant crop damage occurred in SA, NSW and northern Vic. The high levels of persisted adults and hatched nymphs, along with an unknown amount of eggs, threatened to produce an extraordinarily large spring population.

With the observations made by two IMRs, it is clear that the locust plague developed during the season of 1999 – 2000 was mainly due to the success of population migrations into favourite habitats. The first move was from the NSW Central West Slopes and Plains to NSW Upper Western and Qld Far Southwest. The Second move was the westwards movement, populations continued to concentrate into the interior, SA Far North. The third move was the return migration, from SA Far North and NSW Upper Western back to SA Upper North and North East and NSW Riverina. Such movements back to agricultural regions after summer and autumn development in the interior may play an important part in the continuous population cycle.

Decline of Outbreak in 2000 – 2001 Winter rain fell in most areas of southeastern Australia in 2000, and caused overwintering eggs to break their diapause and resume development in late winter. Low temperature in August, however, slowed the egg development.

Spring generation: Large-scale hatchings occurred in the Western of NSW, the Northern Agricultural of SA, and the Mallee and North Wimmera of Vic (Figure A-2f). In early October the APLC started aerial spraying in western and southwestern NSW and Primary Industries of South Australia in eastern central SA. A total of 107 band targets had been treated by the end of October, and a total of 214 bands and 117 swarms by the end of November. Swarms from early hatchings were first seen in the Far Northwest of NSW in late October. Light-trap catches were recorded at Fowler Gap, White Cliffs and Thargomindah continuously in late November, indicating large-scale migrations which coincided with the presence of tropical troughs. The IMR at Bourke, C. Metapopulation Dynamics of Plague Locust during 1998-2001 in Eastern Australia 255 however, did not detect such northward migrations, apparently because they were occurring further to the west. The control effort reduced the population density significantly. In addition, dry conditions in early spring may have caused considerable mortality in eastern SA and western NSW. However, widespread rain in mid-October improved the conditions of survival and of fledging of nymphs. Heavy rainfall in the inland in November led to ideal breeding conditions for local and immigrant populations (Figure A-3b-d).

Summer generation: Swarms were seen in the Far Southwest of Qld in early December. Nymphal bands were also found there and in the Far Northwest of NSW during January. Light-trap catches at Thargomindah and Birdsville indicated small-scale migrations probably occurred within the Channel Country of Qld during the mid- to late December. Population levels in the previous infestation areas of SA, NSW and Vic decreased significantly, possibly due to high mortality caused by dry conditions and emigrations that were detected by the IMR at Bourke and the light-traps at White Cliffs and Thargomindah in mid- and late January. No suitable targets of band or swarms were found during December-January aerial survey in the Far Northwest, where numerous to band density nymphs were seen in early summer. The APLC sprayed 16 bands and 21 swarms in the Far North of SA and the Far Southwest of Qld during the second half of January. The hot and dry conditions during most of December and January may have caused substantial additional mortality in the interior (Figure A-3d-k). Rainfall in late January and early February extended over much of the interior (Figure A-3l-n), where more locust immigrations from the southeast were detected by the IMR at Bourke until the end of the summer and egg laying and hatching produced an asynchronous generation in the Far Southwest of Qld and the Far North of SA.

Autumn generation: The continuous dry conditions in the interior during most of February and March restricted population development (Figure A-3o-r). No bands or swarms were formed that were large enough for aerial spraying. IMR detections and light-trap catches at Fowler Gap suggest that there were small-scale migrations from the interior to the agricultural region of NSW in mid- to late March. Population levels slightly increased in other areas after rainfall in late summer and early autumn. There was little rainfall in April, enforced and this accelerated the collapse of the outbreak.

256 Appendices

Figure A-3 Rain Events in Eastern Australia in Spring – Autumn of 2000 – 2001 Rainfall during (a) 01-07/11/00, (b) 08-14/11/00, (c) 15-21/11/00, (d) 22-28/11/00, (e) 07-13/12/00, (f) 14-18/12/00, (g) 19-25/12/00, (h) 26-30/12/00, (i) 01-07/01/01, (j) 08-14/01/01, (k) 15-22/01/01, (l) 23- 27/01/01, (m) 28/01-01/02/01, (n) 02-07/02/01, (o) 13-19/02/01, (p) 23-28/02/01, (q) 01-06/03/01, (r) 07- 13/03/01, (s) 14-19/03/01, and (t) 20-26/03/01. Green dots indicate the stations with rain recorded. Red dots mean either no rain recorded or no observation. Data source: APLC archived grid rainfall data from the Australian Bureau of Meteorology. C. Metapopulation Dynamics of Plague Locust during 1998-2001 in Eastern Australia 257

In 2000 – 2001, the APLC conducted extensive aerial control on the spring generation in SA and NSW, and small-scale treatment of the summer generation in the interior. A total of 368 bands and swarms were sprayed over an area of 1907.8 km2 using 44,627 litres of insecticide. Although the IMR observations indicated continuous locust migrations occurred, the unusual lack of summer rainfall in the arid interior probably contributed significantly to the suppression of the outbreak.

D. Notation and Statistics of Directional Data 259

D. Notation and Statistics of Directional Data

D.1 Direction Description

Azimuth – the length of the arc of the horizon intercepted between a given point and a reference direction, usually north, and measured clockwise from the reference direction.

Bearing – the horizontal direction from one terrestrial point to another; basically synonymous with azimuth. Bearing may often be expressed in true bearing and magnetic bearing, the angular directions in degrees measured clockwise from true north and magnetic north, respectively.

Elevation – the angle between a point in the space and the horizon.

Any point in space can be located with its azimuth angle, elevation angle, and range.

Vector – any quantity that has both magnitude and direction at each point in space, as opposed to a scalar that has magnitude only. Vector may be represented geometrically by an arrow of length proportional to its magnitude, pointing in the assigned direction. Any vector can be represented in terms of its orthogonal components in given coordinates (see below). Wind is a vector.

Wind direction – the direction from which the wind is blowing. It increases clockwise from north when viewed from above. Terms such as northerly, easterly etc. imply meteorological wind directions φMet . It can also be expressed as a vector azimuth

φVector , i.e. the direction towards which the wind is blowing (wind bearing). Terms such as northward, eastward etc. imply wind vector azimuths.

Wind direction can be expressed with its vector components φPolar in two- dimensions of rectangular coordinates; the angle increases anticlockwise from the X- axis pointing to east.

260 Appendices

Direction conversions between different coordinates are shown in Figure A-4. The directions, we are concerned with are in a horizontal plane (azimuth), and are defined with the zero direction to north and the direction is clockwise. In the study of insect migration, not only the direction, but also the magnitude, such as the insect airspeed and the wind speed, is of importance. However, a vector R is often transformed to its x and y components in rectangular coordinates or a length r at an angle φ in polar coordinates (Cartesian system) for the convenience of calculation. The axis X is horizontal, left to right, the Y axis is vertical, bottom to top, and the angle is rotated counterclockwise, in both the rectangular and the polar coordinates. Since the initial direction, such as north of the azimuth, is arbitrary in the rectangular and the polar coordinates, great care must be taken for the coordinate conversion and rotation. If the azimuthal direction of west to east is as the axis X, and the azimuthal direction of south to north is as the axis Y, a vectorial angle φVect can be converted into a polar angle φPolar by

π φ =−φ A.1 Polar2 Vect

Similarly, if using the polar coordinates as the rectangular coordinates, the vectorial components of R are

⎧ur= cos(φPolar ) ⎨ A.2 ⎩vr= sin(φPolar ) where u and v are the rectangular components of r on the axes X and Y, with the relationship of ruv=+22. Thus,

⎧ v ππ arctan( )− ~ ,ifu > 0 ⎪ u 22 φPolar = ⎨ A.3 v ππ3 ⎪arctan( )+π ~ ,ifu < 0 ⎩⎪ u 22 and with the exceptional cases D. Notation and Statistics of Directional Data 261

⎧π if uv=0 and >0 ⎪ 2 ⎪ φ = 3π A.4 Polar ⎨ if uv=0 and <0 ⎪ 2 ⎪ ⎩undetermined, if uv=0 and =0

An angle can be expressed either in degrees, or in radians. The conversions between these two units are

⎧ 180 φ = φ ⎪ degπ rad ⎨ A.5 π ⎪φ = φ ⎩⎪ rad180 deg

φPolar= π/2 - φVector

φMet= π + φVector r

φVector v u= |r| cos(φPolar)

v= |r| sin(φPolar) φPolar W E φMet u

|r|= sqrt(u2 + v2)

φPolar= arctan(v/ u)

S Figure A-4 Conversions between Coordinates

Tailwind (Following wind) – a wind that assists the intended progress of an exposed, moving object.

The tailwind component is directed along the heading, not the course.

Headwind (Opposing wind) – a wind that opposes the intended progress of an exposed, moving object; the opposite of a tailwind.

Crosswind – a wind that has a component perpendicular to the course (or heading) of an exposed, moving object. Any wind except a direct headwind or tailwind is a crosswind.

262 Appendices

Upwind – in the direction from which the wind is blowing, i.e. moving in a headwind.

Downwind – in the direction toward which the wind is blowing, i.e. moving in a tailwind.

Wind shear – the local variation of the wind vector or any of its components in a given direction. It can be cyclonic or anticyclonic according to the rotation of wind direction.

Wind-shift line – a line or narrow zone along which there is an abrupt change of wind direction.

D.2 Directional Data of Insect Movement

Insect movement is often measured as a vector. The insect displacement is measured in an azimuthal angle of 0 to 360°, which is the direction that the insect is travelling towards, and the insect ground speed. Insect orientation, due to the ambiguity headings of radar measurement, is an axial angle between 0 and 180° from the north clockwise. The wind, on the contrary, is defined as a vectorial angle, from which the wind blows, and the wind velocity, in meteorology. The insect displacement is usually the vector addition of the wind and the insect’s airspeed and orientation. However, it is difficult to measure insect airspeed directly. Instead, it can only be deduced from the insect displacement and the wind vectorially (Chen et al. 1989). Figure A-5 shows the major types of insect displacement with wind.

Heading & airspeed

d e e p s

d n u o peed r s g

& n & o k c a irecti r d T Wind (a) (b) (c) (d) (e) (f)

Figure A-5 Major Relationship Types of Insect Displacement with Wind Vector diagrams indicate the possible orientations of downwind, upwind and crosswind. Insect displacement is the vectorial addition of orientation and wind in downwind (c, b) and crosswind (a, f), or the vectorial subtraction of orientation from wind in upwind (d, e). Adapted from Chen et al (1989). D. Notation and Statistics of Directional Data 263

Orientation data are not circular variables (Batschelet 1981). However, angular data can be transformed to and treated as circular data (Mardia 1972).

D.3 Circular statistics

Orientations, displacements and winds are directional quantities involving an angle which extends a full circle, i.e. 2π radians, and repeats over this period (i.e. θ and θ + 2π represent the same direction). Thus, the analysis of directional data differs from that of conventional linear quantities which extend to infinity and do not repeat (Batschelet 1981). The cyclic nature of the angle variable means conventional statistics cannot always be applied, though analogous methods of linear statistics can be used under some conditions, and to directional data that an alternative ‘circular statistics’ has been developed to cater for variables that extend around an angle. Circular statistics has advanced rapidly in the past three decades (Mardia 1972; Fisher 1993; Mardia & Jupp 2000). Aspects of it which are needed for the analyses of this thesis, and circular statistics, which have been implemented as C++ programs, are briefly described here. They are mainly based on those of Mardia (1972).

Mean Direction and Mean Resultant Length

The mean direction θ of a set of angles θ12,,,θθ" n is calculated trigonometrically from the vector mean of the corresponding points on a (unit) circle:

x0 = arctan SC A.6 where

⎧ 1 n ⎪C = ∑cosθi ⎪ n i=1 in=1,..., A.7 ⎨ n ⎪ 1 S = ∑sinθi ⎩⎪ n i=1

A mean resultant length can also be calculated, as

nn22 221 ⎛⎞⎛⎞ R =+=CS⎜⎟⎜⎟∑∑cosθθii + sin i = 1,..., n A.8 n ⎝⎠⎝⎠ii==11 and must have values in the range 01≤ R ≤ .

264 Appendices

Circular Variance and Circular Standard Deviation

The difference between two angles θi and φ is given by

ξii=min(θφπθφ − , 2 −− ( i )) =−−− ππθφ ( i ) A.9

Thus 0 ≤≤ξi π . As a measure of the dispersion of the angle θi about φ , the quality

11nn Din=−=−−=∑∑(1 cosξθφii ) 1 cos( ) 1,..., A.10 nnii==11

has been found to be suitable to measure it, since 1− cosξi is a monotonically increasing function of ξi over the range 0 ≤≤ξi π . Therefore, about the mean direction θ of θi ,

D is minimised and denoted as the circular variance S0 . Substituting Equation A.7,

SR0 =1− A.11

where R is the mean resultant length. So 01≤ S0 ≤ .

Unlike linear algebra, the circular standard deviation s0 is obtained from S0 as

sS00=−2loge (1 − ) A.12

For small S0 , it can be calculated as

sS00= 2 A.13

Median Direction

The median direction ξ0 of circular data has a similar property to a linear median: it is the angle that has equal numbers of measured points on either side of it, i.e.

ξπ00++ ξ2 π 1 ff()dθθ= ()d θθ= A.14 ∫∫ξξπ00+ 2

where f (θ ) is the probability density function of θi , i=1,…,n, and satisfies

ff()ξ00>+ (ξπ ). The median direction is only meaningful on a symmetrically unimodal distribution, and is the axis of symmetry.

A sample median direction M 0 can be estimated by investigating the minimum circular mean deviation, which is a minimum about ξ0 . Estimation steps are as follows: D. Notation and Statistics of Directional Data 265

1) Count frequencies of angle in n classes with equal class-interval hnl= 2/π in the angular range of 0 to 2/π l ;

2) Count the cumulative frequencies cfξ , and record the half value of the total

frequency cf0.5 ;

3) Calculate the cumulative frequencies of the second half period cfξ +π by

subtracting the opposite values cfξ ;

4) Find the nearby cumulative frequencies of cfi and cfi+1 that include cf0.5 ,

and record both the class-angles of θi and θi+1 , and the cumulative

frequencies of cfi and cfi+1 , i=1,…,n;

5) Interpolate linearly the class-angles of θi and θi+1 with the weights of cfi

and cfi+1 to estimate the median direction

cf0.5 − cfi M 0 =θi +×h A.15 cfii+1 − cf

Rayleigh Test The Rayleigh test is a parametric test, which compares the distribution of a particular sample of directional data with a circular-normal distribution of the von Mises unimodal type. A large resultant length R indicates that the population deviates from a uniform distribution.

Given the data is a random sample of n angular values θ1, ...,θn corresponding to the measurements in a two-dimensional plane, their probability density function f ()θ is

1 Hf:()θ =≤≤ 0θπ 2 A.16 0 2π for a uniform distribution, against the alternative

1 Hf:()θ ≠≤≤ 0θπ 2 A.17 1 2π for a non-uniform distribution, usually the unimodal distribution of von Mises. The statistic in the Rayleigh test is the mean resultant vector length R to be compared with

266 Appendices

the critical values Rα for different significance level α : if R ≥ Rα , the null hypothesis of uniformity H0 will be rejected.

For computer programming, approximations can be estimated for the critical value Rα or the P-values. Baas (2000) gave the simple approximations of critical value

3.00 4.61 RR==, A.18 0.05nn 0.01 for sample n ≥15. Wilkie (1983) estimated the more accurate critical values as

⎧ 2.302 0.174 ⎪R0.10 =−2 ⎪ nn ⎪ 30.75 ⎪R0.05 =−2 ⎪ nn ⎪ 3.689 1.558 ⎨Rn0.025 =−2 ≥5 A.19 ⎪ nn ⎪ 4.605 2.999 ⎪R0.01 =−2 ⎪ nn ⎪ 6.91 8.475 7.625 ⎪ R0.001 =−−23 ⎩⎪ nn n as well as a good approximation

PR()exp144(1)(12)5≥= K( ++ n n22 − K −+ n) n ≥ A.20 for the exceedance probability P when an equal or greater R can be obtained from a sample size n of a uniform distribution. For P ≤ 0.01, this formula gives a slight underestimate (Upton & Fingleton 1985). Although this formula is simple and has good accuracy, the square-root operation may cause domain error. Therefore, the approximation

⎧en−K ≥ 50 2 ⎪ PnR()≥= K 2234 A.21 ⎨ −K 2KK−−+− 24 K 132 K 76 K 9 K ⎪en(1+− ) < 50 ⎩ 4nn 288 2 is adopted (Mardia 1972; Fisher 1993). D. Notation and Statistics of Directional Data 267

Kuiper-Stephens K* Test The Kuiper-Stephens K* test (Upton & Fingleton 1985) is a non-parametric test, which compares the observed and theoretical distributions by the maximal deviation of cumulative probability. Therefore, it can either be used to test the randomness, or the goodness-of-fit for a particular distribution.

Given the null hypothesis is a uniform distribution (Equation A.16), the distribution function is

θ Fin(θ )==i 1,..., A.22 i 2π

The empirical distribution is

i Finn(θ )== 1,..., A.23 i n

Thus, the maximal deviations between F()θ and Fn()θ are

⎧ + n ⎪DFFiii=−Max(0,()θθ ()) ⎨ in=1,..., A.24 − n ⎩⎪DFFiii=−Max(0,()θθ ())

So the Kuiper’s statistic

KDDn=+()+− A.25

+ − + − where, D and D are the maxima of Di and Di respectively. By consulting the table (Stephens 1965), the null hypothesis will be rejected when the statistic K is greater than the upper percentage point. A modified statistic

KDDn* =+(+− )( 1 + 0.155 n + 0.24 n) A.26 can be used to reduce intensive table usage.

The Kuiper-Stephens test can be used for the goodness-of-fit test for a von Mises distribution, if the theoretical distribution is calculated from the parameters of a von Mises (see the Watson’s test below). Also, it can be used to test the homogeneity of two samples, if a second sample distribution is used instead of a theoretical distribution. The statistic will be

+− KnnDD1,2=+ 1 2() 1,2 1,2 A.27

268 Appendices

And the modified statistic will be calculated by

* (1++ 0.155nn11 0.24 ) KK1,2=≤+> 1,2 nnnn 1 2,12 1 2 A.28 nn12() n 1+ n 2

Watson-Stephens U 2 Test Cox’s test (Barndorff-Nielsen & Cox 1979) is a very powerful test of the goodness-of-fit for the von Mises model (Batschelet 1981). However, it is too complicated to implement. The simple alternative available is the Watson-Stephens U 2 test (Lockhart & Stephens 1985; Upton & Fingleton 1985). This is also a non- parametric test, which evaluates the sum of squared differences between the observed and theoretical distributions.

Suppose a random sample of θ12,,,θθ" n is drawn from a von Mises distribution, which has mean direction µ0 and concentration k. Its probability density function f (;θ µ0 ,)k is

⎧02≤≤θ π

1 k cos(θµ− 0 ) ⎪ Hf00:()θθµ== f (;,) k e⎨ k > 0 A.29 2()π Ik0 ⎪ ⎩02≤<µ0 π

where I0 ()k is the modified Bessel function of the first kind and order zero. When k=0, the von Mises distribution reduces to a uniform distribution. The test can be done by calculating the statistic of a mean square deviation

22ππ 2 Un2 =−−− F()θ F ()θφφφθ[] F () F ()d() F d() F A.30 ∫∫00{ nn}

where Fn (θ ) is the empirical distribution function of the n values θi , F(θ ) is the von Mises distribution

θ FF()θ == (;θ µφµφ ;) k f (; ,)d k 0 ≤≤θπ 2 A.31 00∫0

If both µ0 and k are unknown, they can be estimated by the maximum likelihood method. The test procedure is as follows (Lockhart & Stephens 1985; Fisher 1993).

The two parameters of a von Mises distribution M (,)µ0 k are calculated from their maximum likelihood estimators D. Notation and Statistics of Directional Data 269

⎪⎧µˆˆ00= x ⎨ A.32 ˆ −1 ⎩⎪kAR= 1 ()

where x0 is the sample mean calculated as Equation A.6, R the mean resultant length

−1 ˆ calculated as A.8, and A1 ()R is the inverse function of Ak1()= R, which is the maximum likelihood estimate of

Ak110()= I () k I () k= R A.33

ˆ where I1()k is the modified Bessel function of first kind and order one. Thus k can be estimated from a table of A()k . However, approximation is also available

⎧ 5 20.65RR++35 R R ≤ ⎪ 6 kˆ ⎨ A.34 ⎪ 1 23R > 0.65 ⎩⎪2(1−−−−−RR ) (1 ) (1 R )

(Mardia 1972; Baas 2000). A reasonably good approximation is given by Best & Fisher (1981)

⎧ 5 2RR++35 R R < 0.53 ⎪ 6 ⎪ ⎪ 0.43 kRˆ =−⎨ 0.4 + 1.39 + 0.53 ≤ R < 0.85 A.35 ⎪ 1− R ⎪ 1 R ≥ 0.85 ⎪ 32 ⎩ RRR−+43 for n >15. Otherwise,

⎧ 2 max(kkˆ − ,0) < 2 ⎪ nkˆ kˆ = ⎨ A.36 (1)nk− 3 ˆ ⎪ k ≥ 2 ⎩⎪ nn3 + is needed to correct the over-estimate of the true value k when n ≤15 .

The confidence intervals for the mean direction can also be estimated approximately by

⎪⎪⎧⎫u 1 ⎡ nn⎤ ˆˆˆα µµθµθ±−+arcsin⎨⎬⎢n (cos 2∑∑ cos 2ii sin 2 sin 2 )⎥ A.37 ⎩⎭⎪⎪nR 2 ⎣ ii==11⎦

270 Appendices

1 where uα is the upper 2 α point of a unit normal distribution, regardless the distribution form (Upton & Fingleton 1985).

The cumulative frequency value zF= ()θ − µˆ , in=1,..., , can be estimated iik approximately by the modified Bessel functions

∞ ⎧ −221 r ⎪Ik0 ()= ∑ (!)( r2 k ) ⎪ r=0 A.38 ⎨ ∞ ⎪ −+121 rp Ikp ( )=+∑ [( r pr )! !] (2 k ) p = 1,2,... ⎩⎪ r=0

where I p (k ) is the modified Bessel function of first kind and order p and can be calculated by the polynomial approximations (Fisher 1993)

246 ⎧Ik0 ( ) 1+++ 3.5156229 t 3.0899424 t 1.2067492 t ⎪ ⎪ ++0.2659732tt81012 0.0360768 + 0.0045813 t A.39 ⎨ −1246 ⎪kIk1( ) 0.5+++ 0.87890594 t 0.51498869 t 0.15084934 t ⎪ 81012 ⎩ ++0.02658733tt 0.00301532 + 0.00032411 t for 03.75≤

1 ⎧ke2 −−−k I( k ) 0.39894228++ 0.01328592 t12 0.00225319 t ⎪ 0 ⎪ −+−0.00157565ttt−−34 0.00916281 0.02057706 − 5 ⎪ ⎪ +−+0.02635537ttt−678 0.01647633−− 0.00392377 ⎨ 1 A.40 2 −−−k 12 ⎪ke I1( k ) 0.39894228−− 0.03988024 t 0.00362018 t ⎪ −3 −45− ⎪ +−0.00163801t 0.01031555tt+ 0.02282967 ⎪ −678−− ⎩ −+−0.02895312ttt 0.01787654 0.00420059

for k ≥ 3.75 where tk= /3.75. Thus,

2(p − 1) Ik()=− I () k I () k p = 2,3,... A.41 pp−−21k p can be used to estimate the distribution function by the following Fourier expansion

∞ 1 ⎡ I p ()sinkpθ ⎤ Fk(;0,)θθ=+⎢ Ik0 () 2∑ ⎥ A.42 2()π Ik0 ⎣ p=0 p ⎦

Therefore, the cumulative frequency zi can be estimated by D. Notation and Statistics of Directional Data 271

⎧ Fk(;0,)θµi ˆ = 0 ⎪ zF=−(θ µπθµˆˆ ) F (2 +−−− ;0, kFˆˆ ) (2 πµθµ ˆˆ ;0, k ) ≤ A.43 iik ⎨ i i ⎪ FkFk(;0,)(;0,)θ − µµˆˆˆˆ+> θµ ˆ ⎩⎪ ii

Sort zi into ascending order zz(1)≤ ... ≤ (n ) , calculate the statistic

nn2 2 2 ⎡⎤21i − ⎛⎞ 1 1 1 Uz=−∑∑⎢⎥()ii − nz⎜⎟ () −+= in1,..., A.44 ii==11⎣⎦2212nn⎝⎠ n

The null hypothesis that sample is from a von Mises distribution will be rejected if U 2 is larger than the tabulated critical value (Lockhart & Stephens 1985), can be linearly interpolated for a proper entry of k

⎛⎞11 ⎛ 11 ⎞ CV−− CV − CV − k << kˆ k A.45 22112⎜⎟ˆ () ⎜ ⎟ ⎝⎠k kkk212 ⎝ ⎠

where CV1 and CV2 are the significance points corresponding to k1 and k2 at some significance level (Upton & Fingleton 1985).

The confidence intervals for the mean direction of a von Mises distribution can be estimated as

⎧ ⎛⎞1242nR2 − P ⎪µµµˆˆˆ±+≤arccos⎜⎟α CS cos sin ⎜⎟RnP43− ⎪ ⎝⎠α ⎨ A.46 ⎛⎞Pα ⎪ 122 ⎪µµµˆˆˆ±−−+>arccos⎜⎟ 1 (1Re )n C cos S sin ⎜⎟R 3 ⎩⎪ ⎝⎠

2 where Pα is the tail probability of upper α point from a χ1 distribution (Upton & Fingleton 1985). A much simpler formula

⎛⎞1 µˆ ± arcsin ⎜⎟u 1α A.47 ⎜⎟2 ⎝⎠nRkˆ

1 where u1 is the upper 2 α point of a unit normal distribution, was given by Fisher 2α (1993). Similar formulae were also given by Baas (2000). The confidence intervals for the concentration parameter of a von Mises distribution, however, are usually wide and

272 Appendices no useful approximate formula is available when k ≤ 4 (Upton & Fingleton 1985). Direct computation is not available without consulting χ 2 table.

The Watson-Stephens U 2 test can also be used to test if the sample is from a uniform distribution, with the null hypothesis of randomness (Equation A.16), and to test two-sample homogeneity by the statistic

2 nn12++ nn 12 2 ⎛⎞ ()nn12+−∑∑ dkk⎜⎟ d 2 kk==11⎝⎠ U1,2 = 3 A.48 nn12() n 1+ n 2

n n where dnnFkkk=−12() 1()θ F 2 ()θ (refer to the Kuiper’s test). The uniformity of insect displacements and orientations can be tested by the Rayleigh test, the Kuiper-Stephens K* test, and the Watson-Stephens U 2 test to see if there is a preferential clustering angle. The goodness-of-fit for a von Mises distribution can also be tested by the Kuiper-Stephens K* test and the Watson-Stephens U 2 test to check if the measured directions have a circular-normal distribution, so does the two- sample comparison of homogeneity.