SCIENTIA

MANU E T MENTE

CLIMATIC AND OCEANOGRAPHIC EFFECTS ON SURVIVAL OF LITTLE PENGUINS IN SOUTHEASTERN AUSTRALIA

A thesis submitted for the degree of Doctor of Philosophy

By Lucia-Marie Ganendran

Applied and Industrial Mathematics Research Group, School of Physical, Environmental and Mathematical Sciences, The University of New South Wales, Australian Defence Force Academy.

December 2017

PLEASE TYPE THE UNIVERSITY OF NEW SOUTH WALES Thesis/Dissertation Sheet

Surname or Family name: Ganendran

First name: Lucia-Marie Other name/s: Billie

Abbreviation for degree as given in the University calendar: PhD

School: School of Physical, Environmental and Mathematical Faculty: UNSW Canberra Sciences

Title: Climatic and Oceanographic Effects on Survival of Little Penguins in Southeastern Australia

Abstract 350 words maximum: (PLEASE TYPE) Climate change can impact on the survival of seabirds. While many studies have investigated the influences of climatic and oceanographic variables on seabird breeding, fewer have been able to capture the processes affecting survival. In this study, I carried out a mark-recapture analysis on a 46-year penguin dataset to study the effects of some climatic and oceanographic variables on the survival of little penguins Eudyptula minor in southeastern Australia. A priori knowledge of the birds' annual cycle and patterns of movement informed my selection of meaningful and biologically sensible variables.

Two age classes of penguins were considered, based on their differing patterns of movement: first-year birds and adult birds in their second and subsequent years of life.

The climatic and oceanographic variables considered in this study were wind strength, sea-surface temperature, east-west sea temperature gradient, air temperature, rainfall, humidity and chlorophyll a concentration. Climatic covariates which affected adult penguins on land had a direct impact on their survival, most likely via physiological stresses caused by climatic extremes. Oceanographic covariates had contrasting effects on first-year and adult penguin survival. Positive effects were most likely due to the movement of nutrients and associated prey into foraging areas as a result of favourable marine conditions, while negative effects were most likely due to a decrease in prey abundance, or physical factors such as changes in sea-surface temperature or turbulent seas, which affected prey availability or foraging behaviour.

The survival probabilities of first-year and adult birds were most strongly associated with different covariates, and at different times during the birds' annual cycle. The effects were not always immediate, with lagged covariates found to affect survival for both age classes. In a broader context, the effects of any single covariate on the distribution and demography of penguins may be correlated with, or masked by, a range of environmental conditions and interactions between covariates in the marine ecosystem. My research offers new insights into processes which affect penguin survival, and ultimately population security. The development of appropriate management and adaptation actions will further contribute to the conservation of seabirds.

Declaration relating to disposition of project thesis/dissertation

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 in part in the University libraries in all forms of media, now or here after known, subject to the provisions of the Copyright Act 1968. I retain all property 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 350 word abstract of my thesis in Dissertation Abstracts International (this is applicable to doctoral theses only).

…………………………………………………………… ……………………………………..……………… ……….……………………...…….… Signature Witness Signature Date

The University recognises that there may be exceptional circumstances requiring restrictions on copying or conditions on use. Requests for restriction for a period of up to 2 years must be made in writing. Requests for a longer period of restriction may be considered in exceptional circumstances and require the approval of the Dean of Graduate Research.

FOR OFFICE USE ONLY Date of completion of requirements for Award:

Certificate of Originality

I hereby declare that this submission is my own work and to the best of my knowledge it contains no materials previously published or written by another person, or substantial pro- portions of material which have been accepted for the award of any other degree or diploma at UNSW or any other educa- tional institution, except where due acknowledgment is made in the thesis. Any contribution made to the research by oth- ers, with whom I have worked at UNSW or elsewhere, is ex- plicitly 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.

Signed ...... Date ......

i Copyright Statement

I hereby grant the University of New South Wales or its agents the right to archive and to make available my thesis or disser- tation in whole or part in the University libraries in all forms of media, now or here after 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 350 word abstract of my thesis in Dissertation 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.

Signed ...... Date ......

ii Authenticity Statement

I certify that the Library deposit digital copy is a direct equiv- alent 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.

Signed ...... Date ......

iii

Abstract

Climate change can impact on the survival of seabirds. While many studies have investigated the influences of climatic and oceanographic variables on seabird breed- ing, fewer have been able to capture the processes affecting survival. In this study, I carried out a mark-recapture analysis on a 46-year penguin dataset to study the effects of some climatic and oceanographic variables on the survival of little pen- guins Eudyptula minor in southeastern Australia. A priori knowledge of the birds’ annual cycle and patterns of movement informed my selection of meaningful and biologically sensible variables.

Two age classes of penguins were considered, based on their differing patterns of movement: first-year birds and adult birds in their second and subsequent years of life.

The climatic and oceanographic variables considered in this study were wind strength, sea-surface temperature, east-west sea temperature gradient, air temperature, rain- fall, humidity and chlorophyll a concentration. Climatic covariates which affected adult penguins on land had a direct impact on their survival, most likely via phys- iological stresses caused by climatic extremes. Oceanographic covariates had con- trasting effects on first-year and adult penguin survival. Positive effects were most likely due to the movement of nutrients and associated prey into foraging areas as a result of favourable marine conditions, while negative effects were most likely due to a decrease in prey abundance, or physical factors such as changes in sea- surface temperature or turbulent seas, which affected prey availability or foraging behaviour.

The survival probabilities of first-year and adult birds were most strongly associated with different covariates, and at different times during the birds’ annual cycle. The effects were not always immediate, with lagged covariates found to affect survival for both age classes. In a broader context, the effects of any single covariate on the distribution and demography of penguins may be correlated with, or masked by, a range of environmental conditions and interactions between covariates in the

v marine ecosystem. My research offers new insights into processes which affect pen- guin survival, and ultimately population security. The development of appropriate management and adaptation actions will further contribute to the conservation of seabirds.

vi Acknowledgements

I will always be grateful to my supervisor, Leesa Sidhu, for her mentorship through- out my program. She has been a wonderful role model with her boundless patience, kindness and encouragement. My PhD experience will always be defined by her support and guidance. Thank you, Leesa, for making these past few years ones that I will always treasure.

Harvi Sidhu and Peter Dann, my two co-supervisors, have been wonderfully sup- portive. Harvi, with his quirky sense of humour and consummate pragmatism, has helped me to maintain perspective and keep the end goal in sight. How wonderful that our paths crossed. Peter’s constant encouragement was a tonic for my spirits, and his commitment to the environment and to animal conservation opened up a world of wonder previously unknown to me; I am eternally grateful for this.

Ted Catchpole and Lynda Chambers complete the team. Ted was Leesa’s super- visor, and my work is built upon his pioneering techniques. I am grateful that he took an interest in my work. Lynda was my go-to climate person, providing me with climate data and valuable insights into climate change and seabirds. It was she who taught me how to remove a very cross, squawking penguin from its burrow, an unforgettable experience.

Thanks go to Isaac Towers for helpful guidance with handling chlorophyll a data, and to Amy Griffin and Robin Robertson for assistance with drawing some of the maps.

A very special thank you to Peter McIntyre who continues to be a tower of friendship and support.

I would like to acknowledge the wider PEMS community for their tremendous sup- port in creating an environment so conducive to research. I would particularly like to thank Annabelle Boag for her kindness, Julie Kesby for her generous assis- tance, Paul McKie and my many fellow students for their sanity-saving support. My sisters-in-cooking-and-baking, Miza, Mona and Rachael, have kept me smiling with their love and friendship.

vii My candidature has been funded by an Australian Postgraduate Award for which I am grateful.

The support and love of my family has meant everything to me. My mum and dad, Tony and Cecilia, have been rocks upon which to lean. To my siblings, Jaci, Frank and Tony, thank you with much love.

Alex, Andrew and Kate, you make my world complete.

In loving memory of my dad, Tony.

viii Contents

Declaration iii

Abstract v

Acknowledgements vii

Chapter 1 Introduction 3

1.1 Background ...... 3

1.2 Research aims ...... 6

1.3 Novelty of work ...... 6

1.4 Thesis overview ...... 8

Chapter 2 Background of study and data 11

2.1 Study species — the little penguin Eudyptula minor ...... 11

2.1.1 Adults ...... 11

2.1.2 Chicks ...... 13

2.2 Study location ...... 13

2.3 Penguin data ...... 15

2.3.1 Background ...... 15

2.3.2 Raw data ...... 17

2.3.3 Disqualifications from the main dataset ...... 21

2.3.4 Procedure for disqualifying birds ...... 26

2.3.5 Final dataset used in present study ...... 27

ix 2.4 Summarising the data ...... 28

2.5 Preliminary analysis of penguin data ...... 29

2.6 Climatic and oceanographic data ...... 33

2.6.1 Ambient air temperature, humidity and rainfall data . . . . 33

2.6.2 Wind data ...... 33

2.6.3 Sea-temperature data ...... 33

2.6.4 Chlorophyll a data ...... 34

2.7 Creating the covariates ...... 34

2.8 Principal Component Analysis ...... 35

2.8.1 Advantages ...... 35

2.8.2 Disadvantages ...... 35

Chapter 3 Methods 37

3.1 Mark-recapture methods ...... 37

3.2 Maximum likelihood methods and model selection ...... 39

3.2.1 Maximum likelihood methods ...... 39

3.2.2 Model selection ...... 39

3.2.3 Overdispersion ...... 40

3.2.4 Assessing the impact of model covariates ...... 41

3.3 Forming the likelihood ...... 42

3.3.1 Definitions and notation ...... 42

3.3.2 Deriving the likelihood — an individual-based approach . . . 43

3.4 Modelling age dependence ...... 46

3.4.1 Background ...... 46

3.4.2 Age structure ...... 47

3.5 Banding effect ...... 49

3.6 Model construction and notation ...... 51

3.6.1 Goodness of fit ...... 53

x 3.7 Programming the models ...... 54

3.7.1 Alternative programming approaches ...... 55

Chapter 4 Temporal variation in survival and assessing population stability for little penguins 57

4.1 Introduction ...... 57

4.2 Modelling time dependence in survival ...... 59

4.2.1 The reference or full model ...... 59

4.2.2 Checking for overdispersion ...... 60

4.3 Variation in time-dependent survival ...... 61

4.4 Population modelling ...... 63

4.5 Discussion ...... 64

4.5.1 Survival and population numbers ...... 65

4.5.2 Other factors affecting survival ...... 67

Chapter 5 Terrestrial climatic variables 71

5.1 Introduction ...... 71

5.2 Data and methods ...... 73

5.2.1 Penguin data ...... 73

5.2.2 Climate data ...... 73

5.2.3 Model covariates ...... 74

5.2.4 Methods ...... 74

5.3 Results ...... 74

5.3.1 Interaction between banding and the best covariate for adult survival ...... 78

5.4 Discussion ...... 79

5.5 Conclusions ...... 82

Chapter 6 Wind strength and sea temperature 85

6.1 Introduction ...... 85

xi 6.2 Data and methods ...... 87

6.2.1 Penguin data ...... 87

6.2.2 Climatic data ...... 87

6.2.3 Model covariates ...... 88

6.2.4 Methods ...... 91

6.3 Results ...... 91

6.3.1 First-year survival ...... 91

6.3.2 Interaction between banding and the best covariate for first- year survival ...... 100

6.3.3 Adult survival ...... 101

6.3.4 Interaction between banding and the best covariate for adult survival ...... 104

6.4 Discussion ...... 104

Chapter 7 Marine productivity and survival of little penguins — implications of a seasonal coastal upwelling 113

7.1 Introduction ...... 113

7.2 The coastal upwelling process and the Bonney Upwelling ...... 115

7.2.1 Chlorophyll a ...... 115

7.2.2 The Bonney Upwelling ...... 116

7.3 Data and methods ...... 116

7.3.1 Penguin data ...... 116

7.3.2 Chlorophyll a data ...... 117

7.3.3 Study regions ...... 117

7.3.4 Model covariates ...... 118

7.3.5 Methods ...... 119

7.4 Results ...... 119

7.4.1 First-year survival ...... 120

7.4.2 Adult survival ...... 121

xii 7.4.3 Interaction between banding and the best covariate for adult survival ...... 123

7.5 Discussion ...... 124

Chapter 8 Summary and future directions 129

8.1 Summary ...... 129

8.2 Future directions ...... 132

Appendix A Deriving and programming the likelihood 135

A.1 Forming the likelihood ...... 135

A.1.1 Sufficient statistics ...... 136

A.1.2 The likelihood ...... 136

A.2 Model fitting ...... 138

Appendix B Survival, breeding success and significant climatic and oceano- graphic covariates 141

B.1 Introduction ...... 141

B.2 Breeding data ...... 142

B.3 Breeding data and recapture histories ...... 144

B.4 Some breeding statistics ...... 145

B.5 Measuring breeding effects ...... 146

B.6 Breeding patterns and climatic and oceanographic covariates — re- sults and discussion ...... 147

Appendix C Covariate and correlation matrices 153

C.1 Air temperature and rainfall ...... 153

C.2 Wind and sea temperature ...... 157

C.3 Chlorophyll-a concentration ...... 161

Appendix D Principal Component Analysis 165

D.1 Advantages and disadvantages ...... 165

xiii D.2 Method and analysis ...... 166

D.3 Using the PCs in survival models ...... 169

D.3.1 Terrestrial covariates (Chapter 5) ...... 170

D.3.2 Wind and sea temperature (Chapter 6) ...... 170

D.3.3 Chlorophyll-a concentration (Chapter 7) ...... 172

Appendix E LOC and SITE Codes used in this study 175

Appendix F COM, METH and STATUS Codes 177

F.1 COM codes ...... 177

F.2 Methods of Encounter Codes (METH) ...... 179

F.3 Status codes ...... 183

References 185

xiv Publications and Conferences

1. L. B. Ganendran, L. A. Sidhu, E. A. Catchpole, L. E. Chambers and P. Dann, 2011, The effect of directional wind components on survival of little penguins Eudyptula minor, ANZIAM J(E), 52, C1012 – C1030. 2. L. B. Ganendran, L. A. Sidhu, E. A. Catchpole, L. E. Chambers and P. Dann, 2015, Effects of ambient air temperature, humidity and rainfall on annual sur- vival of adult little penguins Eudyptula minor in southeastern Australia, In- ternational Journal of Biometeorology, doi 10.1007/50084-015-1119-2 (based on Chapter 5). 3. L. B. Ganendran, L. A. Sidhu, E. A. Catchpole, L. E. Chambers and P. Dann, 2016a, An investigation into the temporal variation of survival of little penguins breeding on Phillip Island, southeastern Australia, in preparation (based on Chapter 4). 4. L. B. Ganendran, L. A. Sidhu, E. A. Catchpole, L. E. Chambers and P. Dann, 2016b, Effects of local wind and ocean temperature on the survival of little penguins in southeastern Australia, in preparation (based on Chapter 6). 5. L. B. Ganendran, L. A. Sidhu, E. A. Catchpole, L. E. Chambers and P. Dann, 2016c, Marine productivity and survival of little penguins in southeastern Australia — implications of a seasonal upwelling, in preparation (based on Chapter 7). 6. Guest speaker, All Colleges Day ACT, University of Canberra, January 2014, Exploring the world of Mathematics. 7. Presentation at Eighth International Penguin Conference, University of Bris- tol, United Kingdom, September 2013, Effect of climate and oceanographic variables on survival of Little Penguins in south-eastern Australia. 8. Presentation at Australian Mathematical Sciences Student Conference, Aus- tralian National University, Canberra, July 2013, Effect of climate and oceano- graphic variables on survival of Little Penguins in south-eastern Australia. Awarded Best Talk – Mathematics of Planet Earth. maths.anu.edu.au/events/amssc

1 9. Public lecture, The Maths of Planet Earth. Questacon National Science and Technology Centre, Canberra, June, 2013, Climate effects on seabird survival: A statistical analysis of Little Penguins in south-eastern Australia. 10. Presentation at Joint ACT–NSW Australian and New Zealand Industrial and Applied Mathematics (ANZIAM) Meeting, Sydney, November 2012, Is seabird survival affected by a changing climate? Awarded Best Student Talk. 11. Guest speaker, The Science Show, ABC Radio, Australia, November 2012, Phillip Island penguins indicate wider ecosystem health. abc.net.au/radionational/programs/scienceshow/phillip-island- penguins-indicate-wider-ecosystem-health/4376540 12. Presentation at Sixteenth Biennial Computational Techniques and Applica- tions Conference (CTAC), Brisbane, September 2012, Is seabird survival af- fected by a changing climate? A mark-recapture analysis of adult Little Pen- guins Eudyptula minor in south-eastern Australia. 13. Three-Minute Thesis Competition, University of New South Wales Australia Finals, September 2012, The curious mix of penguins and statistics. https://www.youtube.com/watch?v=6yS--V2CBBs. Runner-up. 14. Presentation at Eighth Oamaru Penguin Symposium, Oamaru, New Zealand, July 2012, Climate effects on survival and productivity of Little Penguins at Phillip Island, south-eastern Australia.

2 Chapter 1

Introduction

1.1 Background

The impact of weather on seabirds has long been a major field of study with a rich history in scientific literature (sensu Lack [1950], Williamson [1975], Gradinger [1995], Poloczanska et al. [2013]). With global warming profoundly affecting marine environments [Gr´emillet et al., 2009, Doney et al., 2012], an understanding of fac- tors which affect population ecology is central to wildlife management and species conservation [Jenouvrier, 2013].

As upper-trophic marine predators, seabirds are sensitive to long-term climate trends and shorter-term environmental factors [Reid and Croxall, 2001, Croxall et al., 2002, Weimerskirch et al., 2003]. The International Union for the Conserva- tion of Nature (IUCN) found that, of 9, 856 bird species assessed, 35% have traits that render them susceptible to climate change, with seabirds being the most vul- nerable [Jenouvrier, 2013]. Recent changes in climate have been linked to shifts in the demography of seabirds, affecting species abundance, distribution and phenol- ogy [Sydeman et al., 2001, Crick, 2004, Sandvik et al., 2005, Hipfner, 2008]. For example, later arrival and delayed breeding in several species of Antarctic seabirds have been attributed to a decrease in sea-ice extent in eastern Antarctica [Barbraud et al., 2010]. Similarly, while temporary declines had previously been reported in seabird abundance in the California Current [Ainley et al., 1995], more substantial and possibly climate-related declines and lack of recovery of seabird populations were reported in the 1990s [Oedekoven et al., 2001].

Penguins inhabit the temperate and polar regions of the southern hemisphere, and have evolved a number of physiological adaptations to cold conditions and extended periods at sea. Little penguins, the smallest of all extant penguin species, occur across southern Australia and the islands of New Zealand [Dann and Chambers, 2013]. Many studies have been carried out on this species, providing insights into

3 little penguin genetics, ecology and population dynamics, as well as their response to changes in the environment.

Studies of little penguins on Penguin Island in Western Australia have reported on the population size and distribution of the colony [Cannell et al., 2011], diving and foraging strategies [Ropert-Coudert et al., 2003, 2006], breeding [Klomp et al., 1991, Wienecke et al., 2000] and the effects of environmental covariates [Cannell et al., 2012, 2016]. Studies on other colonies in Australia have provided information on population distributions and breeding ecology, for example on Montague Island, New South Wales [Weerheim et al., 2003], southeastern Tasmania [Stevenson and Woehler, 2007] and Bowen Island, southern New South Wales [Fortescue, 1999]. In New Zealand, Dann [1994] and Perriman and Steen [2000] reported on the distri- bution of the colonies, while others have investigated survival [Renner and Davis, 2001, Johannesen et al., 2002b] and breeding biology [Gales, 1985, Bull, 2000, Heber et al., 2008, Agnew et al., 2014, Johannesen et al., 2002a], providing useful compar- isons with Australian colonies. The effects of a range of environmental covariates such as wave height, sea-surface temperature and Southern Oscillation Index on foraging behaviour and breeding success were analysed by Perriman et al. [2000] in New Zealand and Berlincourt and Arnould [2015] in Australia.

In August 1968, the Penguin Study Group of the Victorian Ornithological Research Group began a flipper-banding program of little penguins on Phillip Island, south- eastern Australia [Reilly and Cullen, 1979]. To date, tens of thousands of little penguins on Phillip Island have been marked with one of two types of tag, a flipper band (that ceased being used in the early 2000s) or a passive induction transponder. Regular burrow visits over the past 46 years have resulted in an enormous amount of raw data, making this one of the longer longitudinal studies of a vertebrate.

Earlier studies using these data were based on small sample sizes and used an unsophisticated analysis technique (life-tables), but provided a sound base for future work. For example, Reilly and Cullen [1979] provided early estimates of adult mortality and patterns of death, Reilly and Cullen [1982] reported on the survival and dispersal of chicks after banding, and Dann and Cullen [1990] analysed survival and reproductive patterns. Other work included Harris and Bode [1981], Dann [1992] and more recently Sutherland and Dann [2014] on distribution and population trends, Dann and Norman [2006] on the competition for food and nesting sites during breeding, and Dann et al. [1992] on movements and mortality at sea. Studies on the breeding patterns of little penguins from this colony include Reilly and Cullen [1981], Dann et al. [1995] and Nisbet and Dann [2009]. Other studies investigated the many aspects of little penguin behaviour such as foraging behaviour [Collins et al., 1999, Hoskins et al., 2008, Preston et al., 2008] and nest attendance and

4 parental provisioning behaviour [Chiaradia and Kerry, 1999, Chiaradia and Nisbet, 2006].

More recent studies using data from this colony include Sidhu et al. [2007] which used 36 years of mark-recapture-recovery data of birds flipper-banded as chicks to study age-related survival. Other studies investigated the effects of environmental covariates: Mickelson et al. [1992] and in later studies, Cullen et al. [2009], Sidhu et al. [2012] and Ganendran et al. [2011] linked sea temperature and wind in Bass Strait with breeding variables or survival of birds from this colony. Ropert-Coudert et al. [2009] associated foraging success with the presence of a thermocline in for- aging waters, and Saraux et al. [2016] reported on the negative effects of wind on foraging behaviour.

Thus, as an extension of previous studies, in this thesis I investigated the effects of several climatic and oceanographic covariates on the survival of little penguins on Phillip Island. I used the remarkable 46-year database of mark-recapture data for chicks and adult birds marked with either a flipper band or an injected transponder. The importance of having clean and reliable data in order to avoid biased estimates cannot be overstated and my first priority in this study was to put the data into a form that would enable the identification of birds with different attributes, for example, birds marked with a flipper band or a transponder, and birds marked as chicks or adults of unknown age with either tag. In practice, this required detailed knowledge of how the raw data were collected and stored, and the writing of extensive computer code to consolidate the records for each individual bird. This procedure is detailed in Chapter 2.

There are several important differences between my work and previous studies. Firstly, the identification of birds marked with either a flipper band or transponder has allowed separate survival estimates for banded and transpondered birds over the entire study period of 46 years. I did this by defining a covariate B which was set to be 1 for a banded bird and 0 for a transpondered bird, based on the findings of Dann et al. [2014] (refer to Section 3.6 for a discussion on banding effect). As there are no data for birds marked with both a flipper band and an injected transponder in the main dataset, I was not able to estimate the probability of tag loss, as in Dann et al. [2014]. Secondly, by including birds marked for the first time only in areas frequently visited for data collection, I attempted to ensure that a critical model assumption of homogeneous recapture probabilities for all marked animals was satisfied (see page 32 for model assumptions). However, in reality the assumption of homogeneous recapture may be violated as some birds do not breed every year and may be less likely to be recaptured (refer to Appendix B for a discussion on breeding patterns). Thirdly, I modelled survival of both age classes of

5 birds over time to explore possible trends in survival over the length of the study. I did this by using time as a covariate for survival as part of the model-fitting process, rather than using a separate least-squares regression of the time-varying survival estimates as input data [Catchpole et al., 1999b]. In addition, I used a simple population model to determine if population changes were in line with predicted survival trends.

The research carried out in this thesis will contribute toward the continuing devel- opment of appropriate adaptation actions to help conserve little penguins on Phillip Island, and to an understanding of seabird demography.

1.2 Research aims

The general aim of this thesis was to understand the effects of the variation in climatic and oceanographic covariates on the survival of little penguins.

The specific aims of this study were:

1. to investigate temporal variation in survival of first-year and adult little pen- guins on Phillip Island for banded and transpondered birds, and evaluate contributing factors; 2. to investigate the effects of terrestrial climatic covariates (ambient air tem- perature, rainfall and humidity) during breeding and moult periods on adult little penguin survival; 3. to investigate the effects of seasonal climatic and oceanographic covariates which operate at sea (wind and sea temperature) on first-year and adult little penguin survival in relation to their annual cycle; 4. to investigate how marine productivity might affect first-year and adult little penguin survival by using chlorophyll a concentration as a proxy; and, 5. to investigate the additive and interactive effects of banding with climatic and oceanographic covariates on little penguin survival.

1.3 Novelty of work

The novelty of my work lies in the following areas.

1. The size of the dataset and the length of the time series for marine vertebrates. An important limitation of many studies investigating relationships between survival and environmental variables is the small number of years of data available. The 46 years of this study allowed the detection of significant relationships among focal variables that may not be otherwise obvious in shorter time series, for example, only six [Catchpole et al., 1999b, Peach et al.,

6 1999, Perdeck et al., 2000, Grosbois and Thompson, 2005, Jenouvrier et al., 2005a, Altwegg et al., 2006] of 78 environmental studies reviewed in Grosbois et al. [2008] had more than 30 years of data. 2. The analysis and handling of a large dataset of little penguin mark-recapture data. 3. The separate parameter estimates for banded and transpondered birds. 4. Bass Strait, the predominant area in which little penguins from Phillip Island forage, is predicted to be a global sea surface temperature hot spot (Figure 1.1, Ridgway [2007], see Figure 2.3 for location of Bass Strait). This is the best dataset to use to investigate the responses of a marine vertebrate to climate variables in southeastern Australia. Penguins are invaluable in predicting responses of ecosystems to climate change [Boersma, 1978, 2008, Forcada and Trathan, 2009, Lynch et al., 2012, Le Bohec et al., 2013]. 5. This is the first time that a detailed analysis has been undertaken of the as- sociation between terrestrial climatic covariates and oceanographic covariates and little penguin survival. Furthermore most of these covariates have not previously been examined in the context of penguin survival. 6. The use of seasonal rather than annual climatic covariates allowed the con- nection of survival with important annual cycle events such as breeding and moult [Grosbois et al., 2006, Nevoux and Barbraud, 2006, Hovinen et al., 2014].

Figure 1.1: The long-term trend in global sea-surface temperature (◦C per century) [Ridgway, 2007].

7 1.4 Thesis overview

In Chapter 2, I present the penguin data and the rationale behind the disqualifica- tion of certain records from the raw dataset. I explain how the final penguin dataset used in my analyses was obtained. The construction of the final dataset represents a significant breakthrough in my work. I also discuss the climatic and oceano- graphic data used in this thesis and a Principal Component Analysis to explore the correlations among covariates.

Chapter 3 describes the methodology used in my analysis and includes the devel- opment of the likelihood and sufficient statistics, as well as an explanation of the model fitting, including banding effects. I illustrate the creation of the sufficient statistics and the likelihood equation for sample data.

In Chapter 4, I begin my investigation into the survival of first-year and adult penguins by modelling time dependence (as well as banding effect) for each of the age classes. I test for temporal trends in first-year and adult survival and use a simple population model to confirm observed survival trends.

Chapters 5 – 7 have been written as stand-alone chapters with appropriate refer- ences to methodology in other parts of the thesis to avoid repetition.

Chapter 5 investigates the effects of terrestrial covariates (ambient air temperature, humidity and rainfall) on adult survival. It was discovered that increasing humidity during moult had a positive association with adult survival, particularly for banded birds.

Chapter 6 investigates the effects of wind and sea temperature on survival. It was discovered that increasing east-west winds in the previous spring had a posi- tive association with first-year survival, particularly for transpondered birds, and the temperature gradient across Bass Strait in the previous winter had a positive association with adult survival, particularly for transpondered birds.

Chapter 7 investigates the effects of marine productivity, using chlorophyll a concen- tration as a proxy, with particular reference to the influence of the Bonney Upwelling in southeastern Australia. It was discovered that the chlorophyll a concentration in the area of the Bonney Upwelling in the summer had a quadratic association with first-year survival, and chlorophyll a concentration in the waters immediately sur- rounding Phillip Island in the summer had a positive linear association with adult survival. The effects were significant for both banded and transpondered adult birds.

Chapter 8 contains conclusions and future work.

8 Appendix A contains a derivation of the likelihood using a cohort-based approach.

Appendix B contains a preliminary study on the interactive effects of breeding with the significant covariates from Chapters 5 – 7 on little penguin survival.

Appendix C contains correlation coefficients and p-values for all the covariates used in this thesis.

Appendix D contains a Principal Component Analysis which tests if the large num- ber of covariates used in this thesis could be meaningfully reduced to ‘synthetic’ linear variables for use in survival models.

Appendices E and F contain, respectively, a description of codes used to denote methods of encounter and the status of penguins, and the locations on Phillip Island used in this thesis.

9

Chapter 2

Background of study and data

In this chapter I present the data used in my thesis; penguin data and climatic and oceanographic data. I show how the penguin data were organised to enable the identification of different attributes such as age and whether a bird was flipper-banded or transpondered. I discuss the use of Principal Component Analysis as an approach in handling large numbers of covariates and possible correlations between the climatic and oceanographic covariates. Some parts of this chapter have been adapted from Ganendran [2011].

2.1 Study species — the little penguin Eudyptula minor

Little penguins occur across southern Australia and the islands of New Zealand [Dann, 2015]. The little penguin is the smallest of all extant penguin species stand- ing at around 33 cm tall. The mean mass of male adult birds is 1,100 g and that of female adult birds is 1,000 g [Dann et al., 1995].

2.1.1 Adults

Adult little penguins are faced with two annually occurring major events throughout their lives. These are breeding and moult, both of which are energetically stressful. Moult takes place between February and April each year and lasts for up to three weeks, during which time the birds are restricted to land. At the beginning of moult, little penguins increase their body mass by up to 40%, and rates of oxygen consumption are significantly increased [Baudinette et al., 1986]. Up to half their body weight can be lost during moult, with starvation and heat stress factors causing mortality during moult fasts [Reilly and Cullen, 1983, Dann, 1992].

The breeding season of little penguins may start as early as May and extend as late as March, with chick-rearing peaking in the late spring and summer months (November – January). During this time, adult little penguins are either on land

11 courting, nest building, incubating or brooding, or they are foraging at sea. This is the most energetically expensive period for adult little penguins, accounting for 31% of their total yearly energy budget in only 16% of the time [Gales et al., 1990]. Egg-laying usually takes place between August and December, with two eggs per clutch; a second clutch may follow a successful first attempt. Incubation of the eggs lasts from 31 to 40 days, with both parents taking it in turn to guard the eggs. Parents left for too long on ‘guard duty’ will desert the nest in search of food, sacrificing breeding in favour of survival [Reilly and Cullen, 1981].

Towards the end of the egg incubation period, foraging trips are shortened and once the eggs hatch, the ‘chick-guard’ period begins, lasting for an average of 15.5 days [Chiaradia and Kerry, 1999]. During this time foraging trips generally only last for a single day, with one parent remaining at the nest all day while the other forages locally and returns in the evening to feed the newly hatched chicks [Reilly and Cullen, 1981]. After the chick-guard period, both parents forage during the day and may not necessarily return at night if foraging has been poor. With both parents away, chick survival may be compromised; fledging success was found to be lowest when parents had to travel further to forage [Collins et al., 1999]. Figure 2.1 shows a chick found dead, possibly due to starvation, outside a nesting box. Dark grey waterproof feathers and brown downy feathers are clearly visible.

Breeding success can generally be predicted by the laying date and the chick weight, with an earlier laying date, increased chick weight and higher number of chicks per pair being associated with a more successful breeding season [Reilly and Cullen, 1981, Chambers, 2004b]. Figure 2.2 shows a male adult with a newly hatched chick and an egg in a nesting box at a study site.

Figure 2.1: Dead chick found outside nesting box at study site.

12 Figure 2.2: Adult bird with newly hatched chick and egg in nesting box at study site.

2.1.2 Chicks

Fledging is a hazardous event as this is when chicks first go to sea to find food. At fledging the immune system is underdeveloped so parasitic infections are more com- mon [Harrigan, 1992]. Newly fledged chicks disperse rapidly from the colony, gen- erally in a westward direction [Reilly and Cullen, 1982, Dann et al., 1992, Weavers, 1992]. The months immediately following fledging are a period of high mortality for chicks [Dann et al., 1992]. After fledging, young birds spend most of their time away from the colony, returning to breed at two or three years of age. From then on they breed every year [Reilly and Cullen, 1981, Dann and Cullen, 1990, Dann et al., 2005].

2.2 Study location

The little penguin colony studied is on Phillip Island, Victoria, in southeastern Australia (38◦300S, 145◦100E, Figure 2.3). Phillip Island lies in the north of Bass Strait, a broad and relatively shallow sea-shelf which is influenced by the confluence of water masses in the region [Sandery, 2007], all of which vary greatly in nutrient content.

The colony is part of the iconic ‘Penguin Parade’ (‘Parade’ hereafter) which is a vital part of the tourist economy of the State of Victoria, generating around $125 million from the nearly half a million visitors it attracts each year. Visitor numbers continue to grow, with numbers increasing by 7.3% in 2012–13 [Ernst and Young, 2012]. Penguin numbers increased from 1984 to 2011 because of conservation efforts [Schumann et al., 2014], but now appear to be decreasing slowly, the reasons for

13 Figure 2.3: Location map of Phillip Island. Courtesy of P Dann. which are not yet fully known. The conservation of these little penguins remains a critical task for the Phillip Island Nature Parks (PINP) [Phillip Island Nature Parks, 2011] and research conducted at PINP provides information to guide wildlife management programs which have regional and international applications.

The distribution of little penguin breeding colonies on Phillip Island as a whole has diminished substantially since the turn of the 20th century, and nine of the ten colonies have now ceased to exist. The remaining colony of around 32,000 individuals is restricted to the main study area, Summerland Peninsula [Dann, 1992]. Loss of breeding habitat on the island was due to human settlement and the accompanying rise in agricultural activity, as well as the introduction of predators and weeds, and the erosion of breeding sites.

In the 1920s, enterprising Phillip Island residents began to take tourists to see the nightly arrival of penguins on Summerland Beach by torchlight. By late 1927, a road and golf course had been constructed and Summerland Estate, a residential development containing 774 housing allotments, was created. In recognition of the impact that the housing estate was having on the penguins, a reserve was established over a small part of the penguin colony (that was later to make up the Parade) in 1955, and in 1968 the Penguin Study Group was established.

In 1985 the Victorian Government adopted the Penguin Protection Plan which included a scientific research and management plan and, remarkably, the buy back of Summerland Estate, which was completed in 2011. A program of revegetation and rehabilitation of the penguin habitat, as well as pest animal management is continuing. More information on PINP including Annual Reports and Management Plans can be found at www.penguins.org.au.

14 2.3 Penguin data

Mark-recapture data for little penguins were first collected on Phillip Island in 1968, and data collection has been undertaken continuously to the present day, making this database on the life histories of a single species of marine vertebrate a rare and valuable one. A review of 78 papers published in the ecological literature found that only six covered more than thirty years of encounter data [Grosbois et al., 2008]. Mark-recapture data used in this study were collected from Phillip Island over a 46- year period from 1968 to 2013. Here I highlight the complexities of these data, their reorganisation and the rationale behind decisions taken to exclude certain records or groups of birds from my analysis. An important aspect of my work was creating two subsets of the dataset — one for penguins that had been flipper-banded and one for those that had had a transponder inserted. This enabled me to obtain separate estimates of the survival probability for banded and transpondered birds, and paves the way for further work on possible interactions between flipper banding and environmental factors.

‘Mark’ or ‘capture’ refers to the first time a bird is encountered, when it is marked with an identifying tag, either a numbered metal flipper band or a passive induc- tion transponder, often referred to as a ‘PIT tag’ (‘transponder’ henceforth). A subsequent encounter is a ‘recapture’ (or ‘re-trap’), and an encounter with a dead bird or the return of a flipper band to the Australian Bird and Bat Banding Scheme (ABBBS; for more information see www.environment.gov.au/science/bird-and -bat-banding) by research staff or members of the general public is a ‘recovery’. ‘Tag’ is a general term which refers to either a flipper band or a transponder. Al- though the inclusion of recovery data could result in more-realistic estimates of survival probabilities when recapture and recovery data are available on the same individuals [Catchpole et al., 1998], I did not use recovery data in my analyses for two reasons. Firstly, recovery data from dead transpondered birds are rarely avail- able since a transponder cannot be read or removed and returned to the ABBBS by members of the public, unlike the majority of the recoveries of banded birds. Secondly, in an earlier analysis of the little penguin data on Phillip Island, survival estimates obtained using mark-recapture-recovery and recapture-only data were almost identical, with no improvement in the standard error when the recovery information was added [Sidhu et al., 2007].

2.3.1 Background

The original study sites on Summerland Peninsula, Phillip Island consisted of 55 burrows, most of which were artificial nesting boxes. Visits to these burrows were weekly for the first three years, and at four-weekly intervals from 1970 to 1988.

15 From July 1988, burrows were checked fortnightly during the breeding season and monthly outside the breeding season. New sites with mainly natural burrows were added in 1982 and were visited during the day, fortnightly, only during the breeding season, from August to April. The numbers of burrows visited have varied over the years, between 130 and 220 burrows in total. In 1986, sites with natural burrows and sites with artificial burrows (170 boxes) were added to the main study area (Figure 2.4) [Dann and Cullen, 1990]. A ‘location’ (or LOC) is an area containing one or more ‘sites’, which in turn contain multiple burrows. For example, Location 1 (Parade, Figure 2.4, Table 2.1) contains seven sites, only five of which were included in this study (see Table 2.2 for frequency of visits and Appendix A for a full list of Phillip Island locations and sites used in this study).

Figure 2.4: Study area used in my research showing the main study locations and sites on Summerland Peninsula, Phillip Island. Courtesy of P Dann.

Table 2.1: Phillip Island locations and sites.

LOC Code Location Sites 1 Penguin Parade 1, 15, 30, 31, 32 3 Cliffs to west 3, 14, 26 4 North Shore 21, 22, 24, 25, 27 28 Summerland Estate North 28 29 Summerland Estate South 29

16 Table 2.2: Frequency of visits to study locations and sites.

Location Frequency Penguin Parade Site 1 was checked fortnightly all year. Site 15 was checked fortnightly from Au- gust to April. The rest of the sites were checked monthly all year. Cliffs to west Checked fortnightly from August to April. North Shore Checked fortnightly from August to April. Summerland Estate North Checked fortnightly all year. Summerland Estate South Checked fortnightly all year.

2.3.2 Raw data

Raw data were obtained from PINP in Microsoft Excel (2010) files. Each row of the raw database is referred to as a ‘record’ and represents a single encounter with a live or dead bird. A bird may be seen only once, that is it is marked for the first time and never encountered again, and so will only have one record. Birds which are encountered on several occasions may have multiple records spanning several years. Each record has 21 columns, each column corresponding to one of the 21 fields shown in Table 2.3.

Little penguins on Phillip Island are either banded as chicks (SX = 3 in the initial record, N = 1) or as adults of unknown age. An encounter with a bird is classified according to the METH code. The first two digits of this code indicate the method of encounter and the second two digits indicate the bird’s status. For example, a METH entry of 2516 indicates a bird that was found sick or injured, was rehabili- tated and released alive with its original band. Birds which are recovered dead will have death-code entries in the METH column of their final record. Birds which are seen on several occasions, then not encountered again, will have zero entries in the METH and COM columns in their final records.

Table 2.4 is a sample of records for three birds from the main raw database. The first bird, with band number BAND = 006827, is an example of a bird that was ultimately recovered dead. It was a female (SX = 2) of unknown age and was encountered for the first time (N = 1) on 11 January, 1975 at study site SITE = 1, the Penguin Parade on Phillip Island (LOC = 1), when it was marked with a flipper band. It was seen alive on four further occasions in the 1975/1976 breeding season, and seen for the last time on 19 July, 1976. On that final occasion, METH code 8405 was recorded, indicating that it was taken by a wild mammal, was dead and the band was removed. This bird was seen at the same location (LOC = 1) for all its

17 Table 2.3: Fields from main raw database. There is one row for each bird on each occasion it is encountered. Fields marked with a † are applicable only for live encounters. If there are any unbanded chicks in the burrow, measurements are taken for fields marked with a *.

DA Date of encounter. DN Day/Night (‘1’ for day- and ‘2’ for night-time encounter, ‘0’ un- known). LOC Location code for live or dead encounters (refer to Appendix A). SITE† Site number (refer to Appendix A). BURR† Burrow number. CONT† Burrow contents (for example, an entry of ‘102’ means that the burrow contains one adult, no eggs and two chicks). BAND Band number. Blank entry for a bird without a band. TRANSP Transponder number for a bird fitted with a transponder. Blank entry for a bird without a transponder. N N=‘1’ for the original banding record of an individual, ‘2’ to indicate an encounter during which a flipper-banded bird was also given a transponder, and ‘0’ otherwise. BD† Bill depth. WT† Weight. HL† Head length. SX† Sex of bird (‘3’ for chick, ‘0’ for unknown, ‘1’ for male and ‘2’ for female). MO† Stage of moult 1–5. METH Australian Bird and Bat Banding Scheme (ABBBS) method of en- counter and status codes (Appendix B). CHW1†∗ Chick weight. CBL1†∗ Chick bill length. CHL1†∗ Chick head length. CHS1†∗ Chick stage 1–5 (whether ready to fledge). COM Comment such as experimental code (Appendix B). EA Location code for oiled birds. encounters but at a different site (SITE = 30) for its final two encounters. It was encountered in the same burrow (BURR = 9116) for the first two encounters, in two different burrows for the next two encounters (BURR = 9097 and BURR = 9022); the burrow number was not recorded subsequently.

The second bird in Table 2.4, BAND = 098231, transponder TRAN = 1379192 and TRAN = 12AE498, is an example of a bird being accidentally transpondered twice. This female (SX = 2) of unknown age was marked for the first time (N = 1) at Cowrie Beach (SITE = 27) at the North Shore (LOC = 4) on the evening (DN = 2) of 7 January, 1998. This bird was in fact part of a study on the effects of banding [Dann et al., 2014], with its first record including a method code METH = 0845

18 Table 2.4: Sample of records from main raw database.

DA DN LOC SITE BURR CONT BAND TRANSP N BD WT HL SX MO METH CHW1 CBL1 CHL1 CHS1 COM EA 11/01/1975 0 1 1 9116 100 006827 0 1 121 1000 0 2 0 0 0 0 0 0 0 0 08/11/1975 1 1 1 9116 200 006827 0 0 120 975 0 2 0 0 0 0 0 0 0 0 06/12/1975 1 1 1 9097 200 006827 0 0 0 950 0 2 0 0 0 0 0 0 0 0 02/01/1976 2 1 1 9022 100 006827 0 0 0 0 0 2 0 0 0 0 0 0 0 0 24/04/1976 2 1 30 0 100 006827 0 0 0 0 0 2 0 0 0 0 0 0 0 0 19/07/1976 1 1 30 0 0 006827 0 0 0 0 0 2 0 8405 0 0 0 0 0 0 07/01/1998 2 4 27 0 100 098231 1379192 1 124 1000 0 2 0 0845 0 0 0 0 413 0 26/11/1998 2 4 27 0 100 098231 12AE498 2 0 1440 0 2 0 0845 0 0 0 0 413 0 01/12/1999 2 4 27 0 100 098231 12AE498 0 0 1160 0 2 0 0845 0 0 0 0 413 0 17/09/2001 1 133 0 0 100 098231 1379192 0 0 0 0 2 0 5405 0 0 0 0 213 0 17/07/2003 1 1 1 8005 200 0 63514EC 1 142 970 0 1 0 0 0 0 0 0 0 0 19 15/11/2003 1 1 31 8000 100 0 63514EC 0 0 950 0 1 0 0 0 0 0 0 0 0 25/11/2003 1 1 1 8005 200 0 63514EC 0 0 1130 0 1 0 0 0 0 0 0 0 0 13/12/2003 1 1 1 8005 110 0 63514EC 0 0 950 0 1 0 0 0 0 0 0 0 0 13/12/2003 2 1 1 8005 210 0 63514EC 0 0 0 0 1 0 0 0 0 0 0 0 0 10/01/2004 1 1 1 8005 100 0 63514EC 0 0 1000 0 1 0 0 0 0 0 0 0 0 10/01/2004 2 1 1 8005 200 0 63514EC 0 0 0 0 1 0 0 0 0 0 0 0 0 16/10/2004 1 1 1 8005 200 0 63514EC 0 0 1100 0 1 0 0 0 0 0 0 0 0 26/05/2007 1 1 32 8064 100 0 63514EC 0 0 960 0 1 0 0 0 0 0 0 0 0 indicating that it was trapped by hand or with a hand-held net and released alive with a band (BAND = 098231) and an electronic tag (TRAN = 1379192).

At its next encounter on 26 November, 1998, at the same location, it was acciden- tally given a second transponder (N = 2, TRAN = 12AE498) and this meant that it was now marked with one flipper band and two transponders. It was disqualified, that is removed from the dataset, with the disqualification code (COM = 413) en- tered into that and the previous record (disqualifications are discussed in detail in Section 2.3.3).

The next encounter with this bird was on 1 December, 1999 at the same location. Only one transponder was detected (TRAN = 12AE498), and again the disqualifi- cation code COM = 413 was entered.

The final encounter with this bird was on 17 September, 2001 at Safety Beach, Port Phillip Bay (LOC = 133, 38◦190S, 144◦590E) when again only one transponder ID was recorded (TRAN = 1379192), the one inserted at the first encounter. The final record has a comment code COM = 213 indicating that it was marked with a transponder plus a band in the transponder experiment. A METH code 5405 in the final record indicates that the bird was found floating in sea or fresh water or washed up on the beach, was dead, and the band was removed.

The third example, a male bird (SX = 1) of unknown age with transponder number TRAN = 63514EC is an example of a bird encountered alive over a span of four years, then never seen again. It was encountered for the first time (N = 1) on 17 July, 2003, when it was marked with a transponder. Its second and final encounters were at different sites (SITE = 31 and SITE = 32) and different burrows (BURR = 8000 and BURR = 8064) but at the same location as all other encounters (LOC = 1, Parade). All other encounters were at the same location (LOC = 1, SITE = 1 and BURR = 8005). All entries in the METH and COM columns are zeros. This bird was encountered for the last time on 26 May, 2007, then not seen again.

Mention should be made here of COM code 112, as its meaning has changed over the length of the study. Before transponders were introduced, birds with a damaged or worn flipper band had the damaged band replaced with a new band, and all records were shifted to the new band number. The records of these birds had METH code 0844, indicating that the bird was trapped by hand or with a hand-held net and was released alive with the original band removed. At PINP, flipper banding ceased altogether in August 2002, and any ‘new’ birds encountered for the first time were marked with a transponder only. In early 2006 any banded bird that was recaptured had its band removed and a transponder inserted. The bird’s records reflected this change of tag, with a COM code 112 (Rebanded) entered on the date

20 of this change. Additionally, all its records were altered to indicate that the bird was transpondered at its first encounter. Thus, at first glance it would seem that a transpondered bird with COM code 112 and METH code 0844 was initially marked with a transponder instead of a flipper band. This potential confusion led to a new status code being introduced in 2013: ‘47 – Released alive, band removed and replaced with a microchip (transponder)’, making it easier to identify such birds. A file of rebanding information such as the original flipper-band numbers, with the new transponder numbers and the dates of change was created to keep track of these birds.

Table 2.5 shows the sample records from the main raw dataset for a rebanded bird and Table 2.6 shows the rebanding information for the same bird. This male bird (SX = 1) was encountered for the first time (N = 1) on 24 May, 1998, when its records show that it was marked with a transponder TRAN = 6B952C8. Its final record (8 November, 2009) shows an entry of METH = 0847 and COM = 112, which indicate that the band was removed and a transponder inserted on this date, although its records indicate at first glance that it was marked for the first time with a transponder on 24 May, 1998. This confusion is clarified when Table 2.6 is examined and it is discovered that the changeover date from flipper band to transponder (CHDATE) was 8 November, 2009; it was in fact first marked with a flipper band (OLDBAND = 099356) on 24 May, 1998 (OLDDATE). All its previous records, as shown in Table 2.5, were adjusted to make it appear that the bird was transpondered on its first encounter on 24 May, 1998. The treatment of rebanded birds is discussed in Section 3.1.3.

2.3.3 Disqualifications from the main dataset

For the results of this study to be meaningful, several subsets of birds had to be disqualified (removed from the dataset). In this section I discuss the groups of bird disqualified and the rationale behind these disqualifications. The following groups of birds were completely disqualified from the main dataset, that is all of their records were disregarded:

– Birds marked for the first time at locations infrequently visited by researchers.

– Birds used for experimental purposes or affected by human intervention.

– Birds that failed to fledge.

– Rebanded birds.

– Birds accidentally given two tags.

– Birds from the final cohort.

21 Table 2.5: Sample of records for a rebanded bird. DA DN LOC SITE BURR CONT BAND TRANSP N BD WT HL SX MO METH CHW1 CBL1 CHL1 CHS1 COM EA 24/05/1998 2 21 0 0 100 0 6B952C8 1 0 1400 0 1 0 0 0 0 0 0 0 0 05/08/2001 2 21 0 0 100 0 6B952C8 0 0 1340 0 1 0 0 0 0 0 0 0 0 08/01/2006 2 21 0 0 100 0 6B952C8 0 0 1300 0 1 0 0 0 0 0 0 0 0 01/04/2007 2 21 0 0 100 0 6B952C8 0 0 1000 0 1 5 0 0 0 0 0 0 0 06/01/2008 2 21 0 0 100 0 6B952C8 0 0 1150 0 1 0 0 0 0 0 0 0 0 02/11/2008 2 21 0 0 100 0 6B952C8 0 0 1250 0 1 0 0 0 0 0 0 0 0

04/01/2009 2 21 0 0 100 0 6B952C8 0 0 1300 0 1 0 0 0 0 0 0 0 0 22 05/04/2009 1 21 0 0 100 0 6B952C8 0 0 1750 0 1 0 0 0 0 0 0 0 0 08/11/2009 2 21 0 0 100 0 6B952C8 0 0 1200 0 1 0 0847 0 0 0 0 112 0

Table 2.6: Rebanded data

OLDBAND OLDDATE AGEBANDED NEWBAND AGE CHDATE LOC NEWTRAN T2ERROR METH CONT SX 099356 24/05/1998 8/11/2009 21 6B952C8 100 1 Furthermore, as mentioned earlier in this section, dead-recovery records were ig- nored.

Birds marked for the first time at locations infrequently visited by re- searchers

Site fidelity of breeding little penguins is high, with large-scale migration of breeding birds between colonies uncommon [Dann, 1992, Peucker et al., 2009, Sidhu et al., 2012]. It is possible for small-scale emigration to occur if birds move to a burrow that lies just outside the study area, although they generally breed in the same burrow from one year to the next. Because of this, ‘apparent survival’ (the product of ‘true survival’ and the probability that the bird does not emigrate) is expected to be very close to ‘true survival’.

Thus it was important not only to include locations with the largest numbers of birds, but to also ensure that the recapture probability remained reasonably consis- tent over the locations. Hence, only birds banded for the first time on Phillip Island at five locations and 15 sites frequently visited for data collection were included in the study. These locations and sites are listed in Table 2.1 (a full list of all Phillip Island locations used in this study is given in Appendix A). These birds were iden- tified by checking the LOC and SITE codes of their record of first encounter (the record containing N = 1).

Birds used for experimental purposes or affected by human intervention

Experimental birds are those used in current or previous studies that may have required interventions that could potentially affect their survival, such as blood sampling or data-logger attachment.

In 1995, around 100 burrows in the main Parade area were used in a separate study. Birds from these burrows were tagged with either a flipper band only or a flipper band and a transponder. The transponders used in this case (‘Tiris’ brand) were different to the ones used by the PINP researchers (‘Trovan’ brand). In the 1995 study, only the last six digits of the seven-digit transponder number were recorded, and these were entered into the main database. Recapture data were not entered except for two years partway through the study. Only the initial mark data for these birds were entered. However, later in the study, as newer computer programs were used, the main database required a seven-digit transponder number and so a zero was added at the start of the six-digit numbers to make them compatible for entry into the new database. Over the course of time, some birds moved into parts of the main study area, and when they were scanned for the transponder number by the ‘Trovan’ scanner, the number appearing on the scanner was a seven-digit number with a ‘true’ first digit, not a zero. Thus this subset of Parade experimental birds

23 might have a flipper-band number and potentially up to two associated transponder numbers. Once identified, these birds and all their records were disqualified from the main dataset.

Between January 1995 and January 2001 an experiment to investigate the effects of flipper banding on adult birds was conducted at Cowrie Beach on Phillip Island [Dann et al., 2014]. Birds were caught at dusk on the beach upon their return from the sea and marked with either one tag (a flipper band or a transponder) or with two tags (both a flipper band and a transponder). Once the study period was over, the burrows in this area were visited infrequently, making the effort associated with data collection inconsistent with that of the main study. Therefore the Cowrie Beach birds were disqualified from the main dataset.

Another study was carried out to investigate the effects of banding on chicks at the Parade, with data collected for chicks that were either flipper banded or transpon- dered (P Dann, unpublished data). This subset of birds could be identified by their LOC and SITE codes, as well as by their COM code, which identified them as experimental birds. These birds were included in my analysis as they were in a location that was frequently visited by researchers and none was double tagged. The birds’ sex given by the SX code (Table 2.3) and their location of initial marking provided sufficient information for inclusion or disqualification from the analysis.

To prevent bias in the survival estimates, penguins whose survival probabilities were thought to be affected significantly by human intervention were removed from the study. These included birds that were involved in experimental procedures, moved to build up a colony in a different location or rehabilitated and released alive. These birds were identified by the appropriate METH and COM codes in their records.

Birds that failed to fledge

Some penguin species such as Ad´eliepenguins Pygoscelis adeliae [Spurr, 1975, Davis, 1982] and king penguins Aptenodytes patagonicus [Le Bohec et al., 2005, Lecomte et al., 2006] display creching behaviour, in which chicks are aggregated within a colony, possibly to reduce rates of heat loss by huddling or as protection against predators. Little penguin chicks however, remain at the nest until fledged, and stay undercover or in the burrows until leaving to go to sea between seven and nine weeks after hatching. If there are two chicks of fledging age, they may leave on the same night but more commonly the chick that hatched first departs two to three days earlier [Stahel and Gales, 1987].

Historically, on Phillip Island, chicks were banded at around six weeks of age, once their flippers were big enough for the band to stay on. The practice of marking a chick at six weeks of age continued when transponders were used. Figure 2.5 shows

24 the guide used by PINP researchers to determine the different stages of a chick’s development from banding to fledging.

Figure 2.5: Guide used to determine banding age of chicks, c1985. Courtesy of PINP.

The classification of the stages of chick moult are: P1: Ready to band (or have a transponder inserted), at around six weeks old. Flipper is completely blue, rest of body still covered in down. 1 P2: Flipper 2 blue or adult plumage. 2 P3: Flipper 3 blue or adult plumage. P4: Flipper all blue, bills smaller, blue brighter and voice squeakier than adults (fully fledged and ready to go to sea).

Since fledging is such a significant event in a chick’s life, the survival of a chick from fledging to around 12 months of age is more meaningful than its survival from marking (which occurs earlier than fledging) to 12 months of age. Furthermore, the birds that died prior to the time of tagging were not recorded in the data set. Hence birds that failed to fledge were disqualified from the analysis. Upon fledging, birds almost always go out to sea, so that a bird was considered to have ‘failed to fledge’ if it was recovered dead in the same location as where it was banded or near the nest. Its records will contain a METH code with the first two digits ‘98’ to indicate this.

Rebanded birds

There are two categories of rebanded birds in the raw database: birds that had their band replaced with another band, usually due to wear or damage of the original band; and birds that had their band replaced with a transponder. Regardless of

25 the reason for rebanding, the contribution to the likelihood of these birds could potentially bias parameter estimates, and so these birds were excluded from the analysis. A total of 659 rebanded birds was excluded.

One option would have been to examine each of the birds individually, include records up to the time of rebanding, then treat the date of rebanding as the final encounter. This could be considered in future work, and is discussed further in Chapter 9.

Birds accidentally given two tags

On very rare occasions, a flipper band was overlooked or missed and a bird was given a transponder, that is the bird was marked with both a flipper band and a transponder. Only four such birds were identified. These four birds were removed from the analysis.

2.3.4 Procedure for disqualifying birds

Raw data were received from PINP with banded birds and transpondered birds in separate sets of files. Each set of files also contained data for any birds with two tags (a flipper band and a transponder). The total number of records I received was 309,176. The earliest record was for 2 February, 1968 for a bird with band number BAND = 000001 (however this bird was disqualified since it was marked for the first time at Melbourne Zoo). The latest record was for 27 June, 2013. Raw data were managed and processed using R [R Development Core Team, 2010]. In addition to disqualifying birds, duplicate entries and data entry errors and discrepancies had to be identified.

Here I present my procedure for disqualifying birds and removing records from the dataset.

1. The most challenging disqualification was the group of experimental birds in the Parade that were marked for the first time in 1995. The marking protocol for these birds meant a bird from this group could potentially have up to three tag numbers associated with it, making it appear to be three different birds. Furthermore, a subset of birds from this experiment was double tagged. Data on tag numbers were provided by PINP which allowed me to remove these birds, but this required a considerable amount of time in checking and re- checking for discrepancies. 4,521 records were removed from the banded files and 5,125 records from the transpondered files, giving a total of 9,646 records corresponding to 2,255 birds being removed from the raw database.

26 2. Next, the experimental birds from the Cowrie Beach banding-effects study were removed by identifying birds marked for the first time (N = 1) at Cowrie Beach (LOC = 4 and SITE = 27) with the experimental COM codes of 13, 113, 213 and 413 (see Appendix B for an explanation of COM codes). A total of 9,460 records was removed. All Parade and Cowrie Beach experimental birds were now removed. 3. Next, I created a subset of birds marked for the first time at Phillip Island locations and sites, as listed in Table 2.1. By using the birds’ first encounter records I was able to obtain BAND or TRAN numbers which enabled me to extract all records associated with those tag numbers. All records associated with a particular bird, and not just its initial tagging record, were removed. This resulted in 138,555 records, corresponding to 38,514 banded birds, and 77,148 records, corresponding to 14,018 transpondered birds marked for the first time at a Phillip Island location and site. 4. Using the subset from (3) above, I created a ‘chick dataset’ using SX = 3 in the record with N = 1 to identify birds that failed to fledge. Since chicks are marked at around six weeks of age, go to sea around two weeks later and bur- row visits are every two weeks during the breeding season, a bird is considered as having failed to fledge if it is recovered dead near the burrows within two weeks of marking. These birds were identified by METH codes 9803, 9805, 9829, 9832 and 9899 in their records (see Appendix B for description of METH and COM codes), and were disqualified. A total of 524 records correspond- ing to 245 chicks which failed to fledge was removed from the dataset (437 records corresponding to 205 banded birds and 87 records corresponding to 40 transpondered birds). 5. Next, I identified and removed all the records of birds with COM codes that indicated human intervention, experimental procedures or scientific studies (see Appendix B for a full list of COM codes used for disqualification), as well as the four flipper-banded birds that had accidentally been tagged with a transponder. 6. The cohort of 1,220 birds that were transpondered in the final year of the study (that is, the 2012/2013 breeding season) were removed from the analysis since there were no recapture data for these birds in subsequent years. However recapture data for this final year were retained. 7. Dead-recovery records were also removed.

2.3.5 Final dataset used in present study

The present study used mark-recapture information from birds marked either as chicks or as adults of unknown age, with records for banded birds ranging from 24

27 February, 1968 to 15 June, 2013, and records for transpondered birds ranging from 16 January, 1997 to 27 June, 2013. The total number of banded birds was 32,936, with 23,355 banded as chicks and 9,581 as adults of unknown age, a total of 107,850 records. The total number of transpondered birds was 11,276, with 7,805 transpon- dered as chicks and 3,471 as adults of unknown age, a total of 70,411 records. In the next section I discuss how these mark-recapture data were summarised into individual life histories.

2.4 Summarising the data

Parts of this section are taken from Ganendran [2011].

Penguin mark-recapture data were summarised into individual life-history vectors for each bird, and used in the construction of the likelihood. Therefore it was crucial to ensure that the summary was an accurate reflection of the penguin’s life cycle. Since major events such as breeding and moult occur annually, it was biologically sensible also to summarise the data annually, rather than monthly or seasonally. Using annually summarised data also had the advantage of minimising the number of model parameters, compared to using monthly or seasonally summarised data.

The vast majority of chicks were marked in the austral summer months of December, January and February, and the majority of initial live captures for unmarked adults occurred at around the same time in the breeding season as the first recapture of marked adults [Sidhu et al., 2007]. Therefore, 1 January was defined to be the nominal census date and ‘birth’ date (i.e. date of hatching).

I defined a ‘Penguin Year’ (PY), where PY tj was measured from 1 July in calendar year tj to 30 June in calendar year tj+1, in order to group together birds hatched or encountered alive in the same breeding season. The following example of life-history data for two birds illustrates how the the data were summarised, and points out the potential loss of information if the data were summarised according to calendar year instead of PY.

Consider the monthly life histories of two birds marked for the first time in the 1969 PY (that is the 1969/1970 breeding season), Bird 1 and Bird 2 in Table 2.7(a). An entry of ‘0’ (denoted by a dot in the table) means that a bird was not encountered during that month and a ‘1’ indicates a live encounter (an initial capture or subsequent recapture). Both birds were banded in the 1969 PY (with nominal census date 1 January, 1970), so that each has a summarised yearly entry of ‘1’ in Table 2.7(b) corresponding to the 1969 PY. Both birds survived and were encountered again in the 1970 PY, and so both have an entry of ‘1’ for the 1970 PY. Neither bird was encountered in the 1971 PY and so both have an entry of

28 ‘0’ for the 1971 PY. Assume that these birds were not seen again after the 1970 PY. Thus, when summarised according to PY, both birds have data entries of ‘1 1 0’. This means that the birds survived the 12 months between their banding and subsequent recapture, that is they survived the 1970 calendar year, after which they were never seen alive again.

Compare this with the same data summarised according to calendar years. Bird 1 would have a ‘0’ entry for the 1969 calendar year, a ‘1’ for the 1970 calendar year and a ‘0’ for the 1971 calendar year. Thus the summarised data entry for Bird 1 would be ‘0 1 0’, indicating that it was first encountered in the 1970 calendar year, and never encountered again after that. However, this bird was encountered for the first time in January 1970 and in fact survived for the whole year, to be seen again in December 1970. This crucial information is lost when summarising data by calendar year. Similarly, Bird 2 would have an entry of ‘1 0 1’. It was not known whether the bird truly survived the second year after marking (calendar year 1971); however the calendar-year method of summary suggests that it had.

2.5 Preliminary analysis of penguin data

I now conduct a preliminary analysis of the summarised penguin data.

Figure 2.7 shows the number of banded and transpondered birds in each cohort (also listed in Table 2.8).

The largest numbers of banded birds were in the cohorts of the decade spanning the mid-1980s to the mid-1990s, dropping off almost completely by 2004, while the numbers of transpondered birds increased significantly in 2002 and have remained high since then. This pattern can be explained by the change in marking protocol of the birds [Dann et al., 2014], from almost exclusive flipper banding until 2002, when birds encountered for the first time were transpondered.

Although 40 birds were marked for the first time with a transponder in the 1997 cohort, three of these birds were experimental birds and the remaining 37 were ‘rebanded’. All 40 birds were disqualified, leaving no transpondered birds for this cohort.

29 Table 2.7: (a) Sample monthly history entries for Bird 1 and Bird 2, and the corresponding annual histories summarised using (b) penguin year and (c) calendar year. For clarity, a zero entry in (a) is denoted by a dot.

(a) 1969 Penguin Year 1970 Penguin Year 1971 Penguin Year 1969 Calendar Year 1970 Calendar Year 1971 Calendar Year J FMAMJJASONDJFMAMJJASONDJFMAMJJASONDJFMAMJ Bird 1 ············ 1 1 ········ 1 1 ··················

Bird 2 ··········· 1 ············ 1 · 1 ··············· 30

Penguin Year Calendar Year (b) (c) 1969 1970 1971 1969 1970 1971 Bird 1 1 1 0 Bird 1 0 1 0 Bird 2 1 1 0 Bird 2 1 0 1 Table 2.8: Numbers of banded and transpondered birds by cohort.

Cohort Banded Transpondered (Penguin Year) 1967 76 - 1968 168 - 1969 511 - 1970 455 - 1971 686 - 1972 398 - 1973 106 - 1974 928 - 1975 460 - 1976 107 - 1977 529 - 1978 308 - 1979 415 - 1980 508 - 1981 1063 - 1982 481 - 1983 350 - 1984 1138 - 1985 2047 - 1986 1449 - 1987 1153 - 1988 1375 - 1989 1606 - 1990 2275 - 1991 1540 - 1992 2181 - 1993 2835 - 1994 2167 - 1995 863 - 1996 974 203 1997 413 0 1998 903 186 1999 787 181 2000 667 162 2001 611 302 2002 339 1154 2003 260 968 2004 1 945 2005 0 1342 2006 0 975 2007 0 1074 2008 0 689 2009 1 1127 2010 0 1046 2011 1 925 2012 0 1220

31 3000

2500

2000

1500

1000 Number of birds banded

500

0 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year

(a) Number of birds banded

1400

1200

1000

800

600

400 Number of birds transpondered

200

0 1996 1998 2000 2002 2004 2006 2008 2010 2012 Year

(b) Number of birds transpondered

Figure 2.7: Numbers of birds banded and transpondered by penguin year.

32 2.6 Climatic and oceanographic data

The climatic and oceanographic data used in this study were generously provided by the Bureau of Meteorology, Victoria, Australia, and the Commonwealth Scien- tific and Industrial Research Organisation (CSIRO), Australia. The data provided were wind speed and direction, sea-surface temperature, ambient air temperature, rainfall, humidity and chlorophyll a concentration. The sources of these data and details on how they were used in the calculation of the covariates used in modelling penguin survival are given in detail in each of the relevant chapters. For complete- ness, I include here a brief description of each of the climatic and oceanographic datasets.

2.6.1 Ambient air temperature, humidity and rainfall data

Data used to calculate maximum daily ambient air temperature and daily rainfall were provided by the Bureau of Meteorology, Melbourne, and taken from the AWAP (Australian Water Availability Project; Jones et al. [2009]) gridded temperature data for the 0.05◦×0.05◦ grid points encompassing Summerland Peninsula (145◦90E, 38◦310S, Figure 2.4), for the period from 1968 to 2013.

Humidity data were obtained from the official database of the Australian Bureau of Meteorology, station number 87031, Laverton RAAF Base (37◦520S, 144◦460E, ele- vation 16 m, Figure 2.3), about 100 km northwest of the study area, as this was the closest site at which humidity data were collected for the entire study period. Hu- midity data from Phillip Island correlated very highly (correlation coefficient = 0.98) with humidity data from Laverton RAAF Base for the days when Phillip Island data were available.

2.6.2 Wind data

Wind data were also obtained from the Laverton RAAF Base. This site was used for several reasons, most importantly because it was the closest site with reliable wind data over the entire study period, and additionally, for consistency with previous studies [Mickelson et al., 1992]. The use of this site was further justified by the fact that adult penguins spend a considerable amount of time in winter and early spring in Port Phillip Bay [McCutcheon et al., 2011], which is relatively close to the Laverton RAAF Base (Figure 1.3).

2.6.3 Sea-temperature data

Sea-temperature data were based on the U.S. National Centers for Environmental Prediction (NCEP) data for the period 1949 – 2001 (after which time the format

33 changed) and corrected satellite data for 2001 – 2013 [Reynolds and Smith, 1994]. The edges of the grid of the sea-surface temperature (SST) data run from 138◦E to 152◦E and from 35◦S to 45◦S, with a 1◦ × 1◦ resolution. For each month there are 10 rows and 14 columns of data, with each data point representing the monthly mean SST at the centre of each grid square. Cells whose centres were on land were ignored. Further explanation of the SST data can be found in Chambers [2004b].

2.6.4 Chlorophyll a data

Satellite-imagery chlorophyll a data were obtained from two sources in order to produce the longest time-series dataset possible. The first was from the SeaWiFS (Sea-viewing Wide Field-of-View Sensor) project for the region enclosed by 35◦S to 40◦S and 138◦E to 148◦E, with spatial resolution of 0.0833◦ latitude by 0.0833◦ longi- tude and a temporal resolution of 7.94 days for the period 1997 − 2010. The second dataset was from the MODIS (Moderate Resolution Imaging Spectroradiometer) project for the same region and with the same temporal resolution but with a spa- tial resolution of 0.0417◦ latitude by 0.0417◦ longitude for the period 2002 − 2013 (data obtained from the NASA Ocean color website, oceancolor/gsfc.nasa.gov and processed by Alistair Hobday, CSIRO).

The two datasets were merged to create a single time series for the period 1998−2012 (see for example Clarke et al. [1988], Cota et al. [2004] and Messi´eand Radenac [2006] for methods), resulting in a 15-year-long dataset. I used the interpolation function interp2 from MATLAB to estimate values for the ‘coarse’ data (MODIS) at the same points as in the ‘fine’ data. In my analysis, I was interested in relative spatial patterns of ocean production rather than absolute values, an approach which has been previously used by Le Bohec et al. [2008] and Boersma et al. [2009].

2.7 Creating the covariates

In creating the covariates, two issues were considered — the time frame and the time lag, if any, to use. When deciding on the former, I needed to consider if a pen- guin’s survival was affected by the monthly variation in a climate or oceanographic variable, or if it was more sensible to use a longer time frame. I chose seasonal rather than monthly covariates for two reasons: first, using seasonal covariates would drastically decrease the number of parameters, so reducing the chances of producing spurious correlations, and second, seasonal covariates were biologically more appropriate since the phenomena which cause mortality for little penguins tend to span months but are less likely to span seasons [Dann et al., 1992].

34 In deciding what time lag to use, I had to consider if an environmental perturbation in the previous year was likely to have an impact on the current survival probability of a penguin. Previous studies have used lagged effects in the study of seabird demography, for example Mickelson et al. [1992], Barbraud and Weimerskirch [2003], Crespin et al. [2006], Cullen et al. [2009]. I chose to allow survival in a calendar year to depend on covariates over six seasons spanning an 18-month period: four seasons in the ‘current’ year and two seasons in the ‘previous’ year.

Each of Chapters 5 – 7 contains a description of the covariates used in survival models and the rationale behind their creation.

2.8 Principal Component Analysis

With the large number of covariates used in my analysis, there was the potential for high correlations among them where more than one covariate might be measuring the same driving principle governing the behaviour of the system.

Principal Component Analysis (PCA) is an approach that can be used to reduce a large set of variables to a smaller set which still contains most of the information of the large set. PCA generates a new set of ‘synthetic’ variables called principal components (PCs), each of which is a linear combination of the original variables. These PCs account for most of the variability in the data.

2.8.1 Advantages

With large datasets, a natural approach is to try to reduce the number of variables but still convey virtually all the information in the original variables. PCA is one of the simplest ways to achieve this. Although this approach is less intuitive than choosing a subset of variables, it has the advantage that a greater reduction in dimensionality can be achieved for the same amount of information loss [Jolliffe, 1990]. Another advantage of PCA is that it allows combining variables of different nature, for example, temperature and rainfall.

2.8.2 Disadvantages

Generally, the PCs with the largest eigenvalues are selected as those contributing most to the total variation. However, variables that could be of biological impor- tance may not be accounted for by these PCs, resulting in the loss of essential information. Thus it is important to select PCs in such a way that all original covariates are well represented; a ‘good’ PC may not be a true reflection of the system under consideration and discarded PCs may be important contributors to

35 the fit of the response variable [Grosbois et al., 2008, Hadi and Ling, 1998, Jolliffe, 1982].

Each PC is a linear combination of the original variables, which can make them difficult to understand and interpret. Additionally, since only linear relationships are considered, this method does not take into account potential higher-order inter- actions between variables [Jolliffe, 1993]. In many applications it may be desirable not only to reduce the number of dimensions but also the number of variables. Mc- Cabe [1984] presents an alternative methodology whereby principal variables are selected, making them easier to deal with than linear combinations of variables.

Appendix C contains the correlation matrices and associated p-values for all co- variates used in this thesis, and Appendix D contains the PCA for these covariates. Appendix D also contains the results from survival models using the PCs as covari- ates and shows that the methods used in this thesis gave better fits of the models to the data and allowed for more useful interpretations of the results.

36 Chapter 3

Methods

This chapter contains methods used in my thesis: included are sec- tions describing mark-recapture and maximum likelihood methods; model selection; the formation of the likelihood; the age structure of the sur- vival and recapture parameters; the model construction; and testing for goodness of fit of a model. I have also included a section on how band- ing potentially affects survival, and a discussion on programming the models using MATLAB.

3.1 Mark-recapture methods

One of the earliest examples of the use of statistics to estimate demographic param- eters was in the seventeenth century. John Graunt (1620 − 1674) used the numbers of dead people recorded in London parishes to estimate and predict mortality due to the plague, eventually creating ‘life tables’ which he used to estimate mortality for different age groups [Hald, 2003]. In 1802, Pierre Simon Laplace (1749 − 1827) estimated the human population of France using a mark-recapture type approach: in his calculations he used a ratio of ‘marked’ individuals (live births) to the total number of individuals in sampled communities [Amstrup et al., 2005]. This funda- mental concept of using ratios between known and unknown values in the study of populations forms the basis of the mark-recapture estimation techniques used today. Landmark work by Cormack [Cormack, 1964], Jolly [Jolly, 1965] and Seber [Seber, 1965] resulted in two models which laid the foundation for current mark-recapture analysis. These are the well known Jolly-Seber (JS) and Cormack-Jolly-Seber (CJS) models. In 1992, a breakthrough paper, Lebreton et al. [1992], reflected the shift away from the emphasis on estimation of population size, to estimating survival rates using mark-recapture models.

In early work on population studies, life tables were used to detect differences in life-history traits such as fecundity and survival over time and space (for example,

37 Deevey [1947], Manly [1977], Clutton-Brock et al. [1985]). However, the analyses of mortality data from life tables contained restrictive assumptions which could not be met by real data, thus rendering biologically unrealistic models and biased pop- ulation parameter estimates [Lakhani and Newton, 1983, Anderson et al., 1985]. A key issue was that age-specific survival estimates were not robust if only young animals were tagged; to overcome this, Brownie et al. [1978] suggested that ani- mals of all age classes be tagged each year. New studies then proposed models for estimating survival rates for animals based on either mark-recovery data or mark- recapture data; see for example Burnham et al. [1987], Pollock et al. [1990], and Freeman and Morgan [1992]. Burnham [1993] proposed combining recapture and recovery data on the same individuals. In 1998, Catchpole et al. [1998] presented a general framework for models that could be fitted to such data while accurately reflecting the main features of the population dynamics of the studied species, and with more-realistic estimates of the survival probability.

The standard scenario for mark-recapture experiments, as defined by Lebreton et al. [1992], is a short mark-recapture occasion with no deaths, births or migrations dur- ing this time, and with deaths (and recoveries) occurring between these occasions. Additional assumptions are: marked individuals are representative of the population of interest; the recapture probability for every marked animal alive at a particular time is the same; the fate or behaviour of marked animals is not affected by the markers; marks are not lost or misread; and the fate of each marked animal is in- dependent of that of other marked animals [Lindberg, 2012]. The above scenario is appropriate for the little penguin data used in this thesis because firstly, birds are mostly seen during the peak breeding season; secondly, most of the deaths occurr once penguins go to sea for extended periods [Dann et al., 1992]; thirdly, migra- tion between breeding colonies during or between breeding seasons is rare [Dann, 1992]. Moreover, the recapture probability remained consistent during this study, as I have included only locations that were frequently visited for data collection (Section 2.3.3). I have also allowed for the effects of banding [Dann et al., 2014] by including a banding variable in the models to indicate whether a bird was banded (rather than transpondered).

More-recent developments have included statistical techniques which added robust- ness to the methods of analysis; these have been driven largely by the rapid ad- vancement in computer technology. Mark-recapture methods now are generally well known and widely practised; a useful introduction for practitioners is provided in Amstrup et al. [2005]. Additionally, reviews such as Lindberg [2012] discuss, among other issues, the consequences of poor study design, and Frederiksen et al. [2014]

38 provide a ‘toolbox for the applied ecologist’, stressing the importance of maintain- ing the collection of high-quality demographic data and more-widespread training in quantitative demographic methods.

3.2 Maximum likelihood methods and model selection

3.2.1 Maximum likelihood methods

By the late 1960s, the use of maximum likelihood methods to estimate unknown parameters of a statistical model was gaining wider practice. This method can be thought of as a measure of the support provided by the data for each possible value of the unknown parameters of interest [Schwarz and Aranson, 1996]. Using likelihood methods in the estimation of unknown parameters in a statistical model begins with the derivation of a function (the likelihood function) which describes the probability distribution of the data, given the model parameters and a specific model form. The next step is to estimate values of the model parameters which maximise the likelihood function, that is the ‘most likely’ values of the parameters [Burnham and Anderson, 2002]. The natural logarithm of the likelihood function is used for ease of computation.

3.2.2 Model selection

When alternative models are proposed for a set of data, the most desirable model is one that adequately represents the data without having more parameters than necessary. Historically, the likelihood ratio test has been the tool for selecting between several ‘biologically plausible’ models [Seber and Schwarz, 2002]. If L1 and

L2 are the maximum values of the likelihood functions under Models 1 and 2, where Model 2 is a more general form of Model 1 (for example, Model 2 is more complex, with v more parameters than Model 1), then the reduction in the log-likelihood in moving from Model 1 to Model 2 is given by  
L1 ∆D = 2 ln(L1) − ln(L2) = 2 ln , L2 where ∆D is the difference in the deviances. If in fact the additional parameters are not needed and have true values of zero, then ∆D will have (asymptotically) a chi-squared distribution with v degrees of freedom. If ∆D is sufficiently large, this suggests that Model 2 is needed to properly describe the data [Amstrup et al., 2005].

While the likelihood ratio test was a useful model selection tool when one of the models under comparison was a more general form of the other (‘nested’ models),

39 different tools were required for non-nested models. One such tool, which can be classified as an ‘information-theoretic’ approach, was developed by Akaike [1973]. Known as the Akaike’s Information Criterion (AIC), it is defined as

AIC = −2 ln(L) + 2K, (3.1) where K is the number of estimable parameters in the model and L is the maximum value of the likelihood function. As more parameters are added to the model, the first term on the right-hand side tends to decrease while the second term increases. This is the trade-off between lack of bias (many parameters) and lack of precision (few parameters) which is fundamental to the principle of parsimony [Burnham and Anderson, 2002].

In practice, the AIC for each candidate model is computed, and the model with the smallest value of the AIC is selected. The AIC is useful for selecting the best model in a set; however, since even the best model might perform poorly in the absolute sense, Burnham and Anderson [2002] stress the importance of ensuring that the set of candidate models is well founded, with biological considerations taken into account.

The AIC is widely used for model selection in biological studies; see for example Grosbois et al. [2008]. Variants to the AIC under certain circumstances have been suggested; for example, when the sample size is small relative to the number of parameters in the model, a second-order AIC (AICc) is recommended. The calcu- lation of AICc uses the equation for AIC (Equation 3.1) with an added correction term for relatively small sample sizes; the correction term becomes negligible for large sample sizes.

3.2.3 Overdispersion

Where overdispersion is suspected, use of a Quasi-AIC (QAIC) is recommended [Burnham and Anderson, 2002, Anderson et al., 1985]. Overdispersion may occur when the underlying model assumption of independence is violated, resulting in more variation (deviance or dispersion) in the data than predicted by the model. For example, lack of independence, heterogeneity or other unknown factors may increase the dispersion of the data.

The calculation of the QAIC also uses the equation for AIC, but with a correction termc ˆ, the estimate of the variance inflation factor (the observed variance divided by the variance under the assumed model, equal to 1 when there is no overdispersion) [Burnham and Anderson, 2002, Grosbois et al., 2008];c ˆ can also be used as a measure of goodness of fit of the model to the data. Withc ˆ incorporated into

40 AIC calculations, the name Quasi-AIC is used to acknowledge the quasi-likelihood nature of the statistic because the data are no longer hypothesised to come from a true likelihood [Amstrup et al., 2005, p.236]. However, it is often not possible to account for some amount of overdispersion in the data [Amstrup et al., 2005, p.19].

There are no ‘general, robust procedures’ for calculatingc ˆ [White and Burnham, 2002], and no one approach works well in every situation. One of the reasons is that, while overdispersion can be accounted for by using the QAIC [Burnham and Anderson, 1998], an accurate calculation ofc ˆ depends on there being a full model that fits the data well. Different approaches to calculatingc ˆ include finding an overall test statistic, χ2, and its associated degrees of freedom, and calculatingc ˆ as cˆ = χ2/df, or using a Monte Carlo approach based on simulation of data [Cooch and White, 2006]. In the Monte Carlo approach a large number of data sets are simulated using parameter estimates from the full model which is assumed correct, as in Catchpole et al. [1999b]. The model is then fitted to each simulated data set and the deviance is calculated each time as in Amstrup et al. [2005, p17]. Finally, the variance inflation factor estimate is calculated as

deviance of the observed data cˆ = . mean of deviances of the simulated data

If the data are overdispersed,c ˆ can be used to scale AIC values, yielding the QAIC, defined as [Amstrup et al., 2005]

−2 ln(L) QAIC = + 2K. cˆ

3.2.4 Assessing the impact of model covariates

The general procedure for assessing the potential impacts of a covariate on survival (such as the climatic and oceanographic covariates used in this thesis) is as fol- lows. The first step is to express the probability of all possible life-histories of the animals in terms of the parameters of interest, such as the survival and recapture probabilities (Section 3.3).

The next step is to construct a reference (or ‘full’) model which just describes the time variation in survival (no climatic or oceanographic covariates are included at this stage, but banding is considered — Section 3.5). Details of the reference model are in Section 4.2.1.

Then I construct models that relate the variation in the parameter, in my case survival, to variation in successive climatic and oceanographic covariates (Section

41 3.6), and measure the statistical support for the effects of these covariates [Amstrup et al., 2005].

3.3 Forming the likelihood

In this chapter, the development of the likelihood is presented using an individual- based approach whereby the likelihood function is formed by taking the product of the likelihood contributions from each individual bird. I have chosen to present an individual-based approach here to illustrate the derivation of the likelihood, as it is much more intuitive than the more sophisticated approach that I have used in my analysis.

My work used a cohort-based approach as in Catchpole et al. [1998], in which the likelihood contribution is taken for each cohort of birds over each mark-recapture occasion, rather than for individual birds. This greatly increased the computational efficiency of my modelling, extremely important given the vast amount of penguin mark-recapture data that were available for analysis. The cohort-based likelihood function is derived informally in Appendix C.

3.3.1 Definitions and notation

The parameters of interest used in this study were the survival and recapture prob- abilities as defined below. Here I present the definitions and notation for the pa- rameters used, in preparation for an informal justification of the individual-based likelihood.

The following is adapted from Sidhu [2007], Catchpole et al. [1998] and Ganendran et al. [2016].

Parameter notation

Let t1, . . . , tk denote the k mark-recapture occasions, and for bird i, let

φi,j = Pr (bird i, alive at tj, survives until tj+1)

pi,j = Pr (bird i, alive at tj+1, is recaptured at tj+1),

for j = ci, . . . , k−1, where ci is the year of first capture of bird i.

Additionally, Catchpole et al. [1998] defined the ‘vanishing probability’, χi,j, as

χi,j = Pr (bird i, alive at tj, not seen again after tj).

42 It should be noted that the above probabilities are conditional on bird i being alive at tj.

In the derivation of χi,j below, the conditional event is omitted for clarity:

χi,j = Pr (bird i not seen after tj)
= Pr bird i dies in (tj, tj+1)

+ Pr (bird i survives till tj+1, not recaptured at tj+1, not seen after tj+1)

= (1 − φi,j) + φi,j(1 − pi,j)χi,j+1, ci ≤ j ≤ k−1.

All values of χ can then be calculated recursively since

χi,k = Pr (bird i not seen after final year of study) = 1.

In other words, the final year of the study is the end of the recapture activity by researchers, and thus the bird will never be seen again.

3.3.2 Deriving the likelihood — an individual-based approach

Here I present an informal derivation of the likelihood using an individual-based approach which is illustrated using an example of four birds over six mark-recapture occasions. Parts of this section appear in Ganendran et al. [2011] and have been adapted from Ganendran [2011].

Recall that the likelihood function, which provides a link between observable (mark- recapture) data and the unknown parameters of interest (survival and recapture probabilities) thought to have produced those data, is the product of the likelihood contributions of individual birds.

Before forming the likelihood, I first create sufficient statistics. These are matrices which form a useful summary of the data, so that the likelihood can be calculated without having to refer back to the raw data. The sufficient statistics are defined as follows:  1 if bird i is recaptured at tj+1, wi,j = 0 otherwise, for j = ci, . . . , k − 1.  1 if bird i is not recaptured at tj+1, but seen alive later zi,j = 0 otherwise, for j = ci, . . . , k − 1.  1 if bird i is recaptured at tj, but not seen again vi,j = 0 otherwise, for j = ci, . . . , k.

43 These matrices have the labels W denoting ‘ones’ (seen alive), Z for ‘zeros’ (not seen) and V for ‘vanished’ [Sidhu, 2007]. Note that wi,j refers to recaptures and therefore does not include the initial capture [Catchpole et al., 1998].

Consider four birds with the following individual life histories from six capture- recapture occasions. A ‘1’ denotes a live capture or recapture and a ‘0’ means that the bird was not encountered on that recapture occasion.

t1 t2 t3 t4 t5 t6 Bird 1 1 1 1 1 0 0 Bird 2 1 0 1 1 0 0 Bird 3 1 0 0 1 0 0 Bird 4 1 1 0 1 0 0

Recall that the two parameters of interest are φ, the survival probability, and p, the recapture probability, and that χ is the probability that a bird, alive at a particular time, is not seen dead or alive after that time. I omit the subscript i (which refers to bird i) to avoid confusion.

Bird 1 survived at least until occasion t4 (with probability φ1 φ2 φ3) after having been recaptured on occasions t2, t3 and t4 (with probability p1 p2 p3). It was not seen again after occasion t4 (with probability χ4). Therefore this bird has a likelihood contribution of φ1 φ2 φ3 p1 p2 p3 χ4.

Bird 2 survived at least until occasion t4 (with probability φ1 φ2 φ3) after having been recaptured on occasions t3 and t4 (with probability p2 p3). It was not seen on
occasion t2 with probability (1−p1) , and was not seen again after occasion t4 (with probability χ4). The likelihood contribution for this bird is φ1 φ2 φ3 (1−p1) p2 p3 χ4.

Bird 3 survived at least until occasion t4 (with probability φ1 φ2 φ3), was not seen
on occasions t2 and t3 with probability (1−p1)(1−p2) , was recaptured on occasion t4 (with probability p3), after which it was not seen again (with probability χ4).

The likelihood contribution for Bird 3 is φ1 φ2 φ3 (1−p1) (1−p2) p3 χ4.

Bird 4 survived at least until occasion t4 (with probability φ1 φ2 φ3), was recaptured on occasions t2 and t4 (with probability p1 p3), was not seen on occasion t3 with
probability (1−p2) , and was not seen again after t4 (with probability χ4). The likelihood contribution for Bird 4 is φ1 φ2 φ3 p1 (1 − p2) p3 χ4.

Below I form the sufficient statistics, matrices W, Z and V for the four birds in the example, using the life histories for these birds together with the definitions of the matrices. W and Z refer to occasions t2 to t6, while V refers to occasions t1 to t6. Each row corresponds to a bird and each column represents an occasion. Note that

44 the first recapture occasion is t2.

      1 1 1 0 0 0 0 0 0 0 0 0 0 1 0 0       0 1 1 0 0 1 0 0 0 0 0 0 0 1 0 0 W =   Z =   V =  . 0 0 1 0 0 1 1 0 0 0 0 0 0 1 0 0       1 0 1 0 0 0 1 0 0 0 0 0 0 1 0 0

By using the entries of row i in each sufficient statistic, I can determine the likeli- hood contribution for Bird i as explained below. I now use an adapted form (for individual birds) of the likelihood equation as developed by Catchpole et al. [2000].

n k−1 k ! Y Y wi,j +zi,j wi,j zi,j Y vi,j L = const × φi,j pi,j (1 − pi,j) χi,j , (3.2) i=1 j=ci j=ci where L is the likelihood, n is the number of birds in the study, ci is the year of first capture of Bird i and k is the number of mark-recapture occasions.

Consider Bird 1 from the example above. I have shown from its mark-recapture data that its likelihood contribution is φ1 φ2 φ3 p1 p2 p3 χ4. I now use the sufficient statistics to construct its likelihood contribution. The sufficient statistics corre- sponding to this bird are contained in Row 1 of each of the matrices W, Z and V above. Now consider the section of the likelihood including the survival parameter,

φ, in Equation 3.2 above. The expression wi,j + zi,j will equal 1 if bird i was re- captured at time tj+1 or if it was not recaptured at tj+1 but seen later. Therefore if wi,j + zi,j is equal to 1, then Bird i is known to survive from tj to tj+1, and so

φj should appear in the likelihood contribution for that bird. In this example, the first row of W + Z is given by (1 1 1 0 0). Therefore the contribution of the survival parameter to the likelihood 5 Y wi,j +zi,j φi,j j=1 is φ1φ2φ3, as before.

Next, consider the recapture parameter, p. Using the first row from W, (1 1 1 0 0), the contribution of 5 Y wi,j pi,j j=1 to the likelihood is p1p2p3. This makes sense as wi,j equals 1 if Bird i is recaptured at tj+1, in which case pi,j should appear in the likelihood contribution.

45 Using the first row from Z, (0 0 0 0 0), the contribution of

5 Y zi,j (1 − pi,j) j=1 is equal to 1, as (1 − pi,j) only appears in the likelihood contribution when the bird is not seen on occasion tj+1, that is when zi,j equals 1.

Finally, consider the final term, χ. Using the first row from V, (0 0 0 1 0 0), the contribution of 6 Y vi,j χi,j j=1 is χ4. Again this makes sense as the ‘vanishing probability’ χi,j only appears if the bird is seen on occasion tj and never seen again, that is when vi,j equals 1.

By multiplying the components corresponding to each of the parameters, the like- lihood contribution of Bird 1 is φ1 φ2 φ3 p1 p2 p3 χ4, as stated above.

The likelihood contribution for the other birds in this example can be derived in a similar manner, and the likelihood for all of the birds in this example is the product of the likelihood contributions of individual birds. Thus I have provided an informal justification of the likelihood equation, Equation 3.2.

Using the method of maximum likelihood, each of the parameters is estimated such that the likelihood is as large as possible. That is, I obtain parameter esti- mates which maximise the likelihood, making the observed results the most prob- able given the model [Manly et al., 2005]. For ease of computation, the natural logarithm of the likelihood (log-likelihood) equation is used; the same parameter values that maximise the log-likelihood will also maximise the likelihood function [Manly et al., 2005]. Rather than computing the parameter values that maximise the log-likelihood, I used the fmincon function in MATLAB to minimise the nega- tive log-likelihood.

3.4 Modelling age dependence

3.4.1 Background

Early studies of seabird survival were empirical, using a life-table analysis with the critical assumption that annual mortality was constant during the life of adult birds [Lack, 1943, 1954, Manly, 1977]. Stochastic models derived by Jolly [1965] and Seber [1965] required assumptions of constant survival and constant recapture.

46 Later models specified time-varying survival [Seber, 1970]. These models also in- cluded time-varying recovery rates, but did not include different age classes of birds, while others included age-dependent rather than time-dependent survival but with constant recovery rates [Seber, 1971]. Further development extended the original Jolly-Seber models to allow survival and recapture rates to vary with an animal’s age [Pollock, 1981]; studies which assumed that all animals in a population, regard- less of their age, shared the same biological parameters, were found to be unrealistic [Pollock, 1981, Barker, 1999]. Even-more-sophisticated models allowed for the in- vestigation of the variation in survival and recapture probabilities with time, age, and other time-specific (such as time since banding) and individual (such as sex) covariates [Lebreton et al., 1992, 1993].

New statistical approaches and the rapid advancement of technology (both hardware and software) permitted the development of robust and flexible techniques which could provide more-precise estimates of survival using mark-recapture data [Am- strup et al., 2005, Manly et al., 2005]. Catchpole et al. [1998] provided a framework for modelling fully age- and time-dependent parameters using a far-more-efficient method that determined the likelihood as the product of the likelihood contribu- tions of each cohort of animals rather than of each individual animal. The statistical techniques used in this thesis are detailed in Section 3.7.

3.4.2 Age structure

It is essential for models to include an age structure in the parameters in order to be biologically realistic and to reflect accurately the population of the animal studied. Unmodelled age variation in one parameter may cause an apparent age dependence in another, referred to as ‘leakage between parameters’ [Catchpole et al., 1998, 2004, Sidhu et al., 2007].

Survival parameter

The penguins used here comprised birds marked as chicks and adults of unknown age. Sidhu et al. [2007] proposed a relatively complicated age structure for survival, with separate survival probabilities in the first, second and third years of life, and a constant annual survival thereafter, declining gradually after nine years of age. However, since this earlier work showed that the most important distinction between the age-dependent survival estimates was between first- and second-year birds, I have chosen to consider a simple survival age structure with distinct annual survival probabilities in the first year and in the subsequent years of life (as in Sidhu et al.

[2012]). This is written as φ1 for the survival probability of first-year birds (birds in their first twelve months of life), and φ2 for birds in their second year and above (referred to as adult survival). See Table 3.1 for the model notation used in Chapters

47 4 – 7. It should be noted that the second age class for survival includes both breeders and pre-breeders that may have come ashore only to moult (approximately 50% of two-year-old birds breed, Dann and Cullen [1990]). It also includes birds marked as chicks two or more years earlier and birds marked as adults of unknown age.

This age structure for the survival probability is a biologically sensible reflection of the different patterns of movement of first-year and adult little penguins. Chicks fledge at around eight weeks of age and go to sea, travelling to distant foraging areas and returning to the colony to moult for the first time after twelve to fourteen months. First-year birds are particularly at risk due to inexperience and internal parasitic loads [Dann and Cullen, 1990, Dann et al., 1992, Harrigan, 1992]. In contrast, once penguins reach breeding age, they generally remain relatively close to the colony during the breeding season. Some make rare longer trips during the non-breeding season, travelling along the coast for up to several hundred kilometres [Dann et al., 1992, Weavers, 1992, Collins et al., 1999]. The distinctly different patterns of movement at sea between first-year and adult birds suggested that these age classes would be affected differently by climatic or oceanographic covariates; thus the choice of age structure for survival is a logical one.

Since there are no existing data to allow estimation of small-scale movement into and out of the colony, I did not include immigration/emigration in my models. Thus, the models I used provide estimates of apparent survival after any movements of birds are taken into account. Apparent survival is defined as φ = s(1 − e), where e is the probability that a bird emigrates and s is the true survival probability. Local emigration may contribute to the underestimation of true survival.

Recapture parameter

Similarly, the age structure for the recapture probability reflects the different pat- terns of movement of little penguins. Based on Sidhu et al. [2012], a biologically sensible age structure was to consider separately birds in their first, second and subsequent years of life, since birds in their first year were rarely seen, birds in their second year were more likely to be recaptured as they returned to the colony to moult or to breed for the first time, and the more experienced breeding birds in their third and subsequent years of life were more likely to spend time at the colony, leading to a increase in the probability of being recaptured. Therefore, there are three age classes for the recapture probability.

I have used an (age + time) model for recapture probability as in Sidhu et al. [2012]. This implies that the time-varying factors affecting recapture affect all three age classes similarly, that is recapture varies in parallel (on a logistic scale) for each age class.

48 3.5 Banding effect

The little penguin dataset used in this thesis contains birds that were either flipper banded or injected with a transponder. Therefore it is important to discuss the potential effect of banding on the survival.

The banding of penguins has a long history and has allowed the accumulation of large amounts of data which have been used to study life-history traits and behaviours of penguins [Austin, 1957, Le Maho et al., 2011], as well as providing insights into observed changes in penguin populations, such as in the assessment of the success or otherwise of the rehabilitation of oiled birds [Nel and Whittington, 2003]. However, studies have found indications of negative effects of banding on body condition, linked to foraging performance, which ultimately contribute to a reduction in reproductive output and survival.

It is possible that the effects may be cumulative and not apparent for several years. For example, longer-term effects can be seen in an increase in the number of forag- ing trips by banded compared with unbanded Ad´eliepenguins Pygoscelis adeliae, possibly exacerbated over the winter breeding period when the birds have to swim greater distances to forage [Dugger et al., 2006]. Culik et al. [1993] found that banded Ad´eliepenguins on Ross Island, Antarctica, incurred significantly higher energy costs at swimming speeds between 1.4 and 2.2 m/s than unbanded penguins (21.1 W/kg compared with 17.0 W/kg, mean power increase was 24% when swim- ming with a flipper band, ANOVA, paired design, n = 7, p = 0.006). This increase in energy costs could potentially lead to a reduction in breeding success especially during periods of low food availability. As a result, banded birds may lose body condition over the breeding period and may not recover sufficiently to breed in the following season or survive the following year [Hindell et al., 1996, Froget et al., 1998].

Differences in breeding parameters and survival between banded and unbanded birds have also been reported. For example, unbanded king penguins Aptenodytes patagonicus arrived earlier for courtship, and had a greater probability of breeding and higher survival than banded king penguins [Gauthier-Clerc et al., 2004]. Dugger et al. [2006] also reported a strong negative banding effect on apparent survival (estimate of banding effect = −0.80, 95% CI = −1.46 to −0.13) resulting in a decrease of 11 to 13% in apparent survival for banded compared with unbanded Ad´eliepenguins. Dann et al. [2014] reported lower survival probabilities in little penguins marked with flipper bands than in those marked with transponders: 74.5% (95% CI = 70.8% to 77.8%) compared with 80.8% (95% CI = 75.1% to 85.4%) in

49 the first year after tagging, and 87.2% (95% CI = 84.2% to 89.7%) compared with 91.4% (95% CI = 87.6% to 94.1%) in subsequent years.

The negative impacts of flipper banding are primarily due to the effect of drag on swimming performance and injury caused by the band, making band type, design and material important considerations [Culik et al., 1994]. Magellanic penguins Spheniscus magellanicus tagged with double bands (aluminium and stainless-steel) showed different effects of banding — aluminium bands caused more feather wear, tissue injury and death than stainless-steel bands. Having two stainless-steel bands had little impact on the survival of male Magellanic penguins, but reduced survival of females by about 8%. However, there was no effect on egg size or reproduc- tive success of females [Boersma and Rebstock, 2010]. African penguins Spheniscus demersus fitted with silicone rubber bands designed to reduce hydrodynamic drag did not behave differently in captivity to unbanded penguins nor was their breeding success any different to steel-banded or unbanded birds in the wild [Barham et al., 2008]. Hampton et al. [2009] also reported no significant difference in breeding success between banded and unbanded African penguins at Robben Island, South Africa. Boersma and Rebstock [2009] found no difference in foraging-trip dura- tion between flipper-banded and web-tagged Magellanic penguins at Punta Tombo, Argentina.

A review of published studies sought to assess the effects of flipper bands on penguin ecology, juvenile and adult return rates to the colony, annual survival, over-winter survival and energetics of swimming [Jackson and Wilson, 2002]. The authors concluded that the use of flipper banding was probably detrimental to the birds’ survival, particularly among juveniles and during times of prey shortage.Petersen et al. [2006] reviewed the growing body of evidence on the negative effects of some flipper bands and many other studies, for example Dugger et al. [2006], Gauthier- Clerc et al. [2004], Fallow et al. [2009] and more recently Boersma and Rebstock [2010], Le Maho et al. [2011], Wilson [2011] and Saraux et al. [2011], have reported detrimental effects of flipper banding.

Of particular interest to my research, Dann et al. [2014] showed that little pen- guins marked with flipper bands had mortality rates about one third greater than transpondered birds, and therefore, included in the models in my thesis is a banded indicator variable B (B = 1 for banded birds, 0 for unbanded birds) to accommo- date the difference in survival for banded and transpondered birds. Dann et al. [2014] also found that survival for banded and transpondered birds was lower in the first year after tagging. An immediate effect of banding was detected in other stud- ies; Fallow et al. [2009] found that upon banding, little penguins altered their diving behaviour and attributed this most probably to the trauma of marking. A similar

50 immediate effect of banding was found in captive Ad´eliepenguins with banded birds swimming at slower speeds than unbanded birds [Culik et al., 1993]. Additionally, the effects of flipper banding may interact with factors such as extreme weather events and food availability which are known to vary from year to year [Saraux et al., 2011]. Therefore I have also modelled the interactive effects of climate and banding on the survival of banded and transpondered little penguins.

Mention should be made of tag-loss probabilities as these can bias survival esti- mates. By using double-tagged birds, Dann et al. [2014] estimated transponder loss probabilities to be higher than flipper-band loss probabilities particularly in the first year after marking (5% compared with 0.7%). As my study only contained birds with one type of tag, I was not able to estimate tag-loss probabilities. However, the inclusion of tag-loss probabilities is suggested for future studies and is discussed in Chapter 9.

3.6 Model construction and notation

I now construct appropriate models for the survival and recapture probabilities to investigate the age, time and covariate dependence of these parameters. I have kept this section brief to provide background, with a more detailed discussion in Appendix C.

First, parameter (survival and recapture) dependence for each age class is set to be time- or covariate-dependent. I now redefine the subscript i to refer to the age class of the survival and recapture probabilities. In this thesis, I used an age structure for the survival probability as defined in Section 3.4.2 in all models; thus i = 1 refers to first-year survival (denoted as φ1) and i = 2 refers to adult survival (denoted as

φ2). The various combinations of time and covariate dependence for the survival age classes, along with the shorthand notation used, are shown in Table 3.1.

51 Table 3.1: Notation and interpretation for survival models.

Notation Meaning

φ1 Probability of survival of birds in their first year of life, that is ‘first-year survival’.

φ2 Probability of annual survival of birds in their second and subsequent years of life, that is ‘adult survival’.

φ1(C) Probability of first-year survival is constant.* φ1(t) Probability of first-year survival varies with time.* φ1(B) Probability of first-year survival depends only on whether or not the bird was banded.* B = 1 for banded birds, B = 0 otherwise. Referred to as the null model.*

φ1(t + B) Probability of first-year survival varies with time and depends on whether or not the bird was banded. Referred to as the full model.*

φ1(X + B) Probability of first-year survival depends on a climatic or oceanographic variable X and on whether or not the bird was banded.*

*Similarly for φ2.

A model with first-year survival depending on a single covariate X and whether or not the bird was banded, B, at the first stage of modelling, was expressed as

logit (φ1,j) = b0 + b1X1,j + cB, (3.3)

th where φ1,j is the probability that a first-year bird survived the j calendar year, given it was alive at the start of that year. The logit function logit (x) =log x/(1 − x)
, the logistic transform, is used as a link function to ensure that the parameter estimates lie in the interval [0,1], and b0, b1 and c are unknown parameters that were estimated using maximum likelihood methods.

At the second and subsequent stages of modelling, additional covariates were added to Equation 3.3, so that at the sth stage of modelling, the fitted model was

logit (φ1,j) = b0 + b1X1,j + ... + bsXs,j + cB, (3.4)

where b0, b1, . . . , bs and c are unknown parameters. Similar models were written for

φ2,j. In this way I was able to consider the interactive effect of age and the covariate on survival.

If a covariate was found to be highly correlated to another covariate in a previous stage, the next-best model (ranked by AIC values) was considered. This approach ensured that covariates of high biological significance were not excluded before the start of modelling.

52 In all models in this thesis, I used an age + time model for the recapture proba- bility pi,j, with a recapture age structure as defined in Section 3.4.2, denoted as p1, p2, p3 (age + time), and modelled as

logit (pi,j) = αi + βj,

where α1, α2 and α3 are constants specific to each recapture age class and the βj are the time-dependent parameters. As mentioned in Section 3.4.2, this means that the recapture probability varied in parallel (on a logistic scale) for the three recapture age classes.

3.6.1 Goodness of fit

There are many statistical tools and techniques available for the analysis of bio- logical data and the study of links between climate variables and the dynamics of natural populations. The choice of technique requires an understanding of the number and quality of data available, and the causal relationships under analysis.

A commonly used measure of the goodness of fit of a model is to compare the amount of deviance explained by the covariate against the amount of deviance of the model when the covariate in question is not included. Nakagawa and Cuthill [2007] suggested the use of a statistic which measures the proportion of deviance explained by the covariate under consideration [McCullagh and Nelder, 1989, Skalski, 1996]. The proportion of deviance is a measure of the amount of the temporal variance in survival that is accounted for by the relationship with the covariate. Therefore, the fit of a model with log-likelihood l is determined by the proportion of total time deviance explained, calculated as (l − l0)/(lfull − l0), where l0 is the log-likelihood of the null model (survival depending only on whether or not the bird was banded) and lfull is the log-likelihood of the full model (survival dependent on time and whether or not the bird was banded) [McCullagh and Nelder, 1989, p.33].

A review of 78 ecological papers published between 1985 and 2006 [Grosbois et al., 2008] evaluated the tools used to assess the effect of climatic factors on the sur- vival of vertebrates through the analysis of mark-recapture data. Three common approaches used in the papers were discussed: the null-hypothesis testing approach; the information-theoretic (IT) approach; and the Bayesian approach. Of the pa- pers reviewed, 33 used an IT approach, with model selection via AIC. A prevalent limitation was datasets with a small number (fewer than 30) of years of animal encounter history data, which potentially produced spurious relationships resulting from an inability to identify a priori a limited set of candidate climatic factors.

53 3.7 Programming the models

The use of statistical methods and software such as MATLAB to consolidate likeli- hood ideas in population dynamic studies has become more widely practised, with a wide variety of model structures available to analyse mark-recapture and recovery data. The development of statistical methodologies is described in, for example, Lebreton et al. [1993], and in this chapter. Here I describe the development of the use of MATLAB for mark-recapture analysis in the context of my research, as well as alternatives programs which are available.

Catchpole [1995] used MATLAB to perform interactive analyses from ring-recovery (recovery) and mark-recapture studies. By first writing the likelihood in terms of the parameters (typically survival and recapture probabilities), different models for survival and recapture could be specified. The parameters of interest and their stan- dard errors could then be estimated using methods of maximum likelihood, making it unnecessary to assume equality of recapture probability (an earlier shortcoming), and goodness-of-fit tests could now be carried out. This paper contained MAT- LAB code and a simple notation and methodology for modelling the dependence of model parameters on external covariates. The high-level MATLAB language made it relatively simple to construct and analyse arbitrary models.

Later, Catchpole et al. [1998] demonstrated a method for integrated analysis of recapture and recovery data which provided full flexibility in terms of the extent of age and time dependence in the parameters of interest. The likelihood, pre- sented for the first time, was constructed with a focus on age variation in survival and recapture probabilities, and considered cohorts of marked animals, rather than individual birds. For a species that requires age- and time-dependence of model parameters, this analysis presented a comprehensive picture which accords with biological knowledge, as demonstrated in that paper.

In 2000, Catchpole et al. [2000] provided a framework for the regression analysis of mark-recapture data and mark-recovery data to investigate the influence of both individual and environmental variables on survival. The likelihood was constructed such that the parameters of interest were allowed to have arbitrary age and time dependence, including the modelling of time dependence via logistic regression on environmental covariates. This allowed for both individual- and cohort-based co- variates to be dealt with in the structure of the likelihood. Their likelihood for the general integrated model provided a flexible framework for model refinement and selection.

My research continues to build upon the methods in Catchpole et al. [1998] and Catchpole et al. [2000], and uses techniques developed over the years at UNSW,

54 together with non-trivial modifications for analysis of the newly summarised raw penguin data. These techniques have been extensively used and published, and provide an environment which, although mathematically intense, offers a great deal of flexibility in approach (for example, Catchpole et al. [1999a,b, 2000], Sidhu et al. [2007, 2012], Dann et al. [2014]). This approach has the advantage of being an integrated process in which the regressions on climate and oceanographic covariates take place as part of the model-fitting process, and not afterwards as a separate least-squares regression on covariates [Catchpole et al., 1999b].

3.7.1 Alternative programming approaches

There is a range of analytical methods used to model data from marked individuals in the study of the dynamics of wild animal populations. A useful review of methods available to the applied ecologist (using data from individually marked animals) is contained in Frederiksen et al. [2014]. Different approaches, such as Bayesian statistics, are used in for example, Dupuis [1995], Gimenez et al. [2007] and Colchero et al. [2012]. However, many of these methods are often time consuming and pose challenges for non-statisticians.

Alternatives to the more mathematically intense approaches lie in the use of avail- able software packages, such as the programs MARK and E-SURGE, which offer user-friendly tools with a WINDOWS interface. Although these are popular choices, they are themselves non-trivial to use and require expert assistance to avoid errors [Lettink and Armstrong, 2003]. A perusal of the detailed documentation that ac- companies MARK indicates that the developer of the program, Gary White, is con- stantly refining and updating the code (warnercnr.colostate.edu/ gwhite/mark/ mark.htm). The theory and methodology behind MARK is provided as a separate attachment to the main program, but is not necessary for those who wish to use the program simply as an analysis tool.

E-SURGE, developed by Choquet et al. [2009], is used for fitting multi-event models with ‘hidden states’ (for example, an event could be ‘not seen’ or ‘seen breeding’, and a hidden state’ could be ‘breeder’ or ‘dead’). The user manual is available at http://occupancyinesurge.wdfiles.com/local-files/start/E-SURGE-MANUAL .pdf and offers a sample ‘session’ for the new user. A degree of expertise is required to understand the complexities of the mathematics behind the program and to set up the models.

In conclusion, writing code for analysis will always be preferred by some, and is a flexible and powerful approach which allows complex models to be built and modified (for example, Brooks et al. [2000], Cubaynes et al. [2014]) without being limited by the constraints or the ‘black-box’ approach of software packages.

55

Chapter 4

Temporal variation in survival and assessing population stability for little penguins

The aim of this chapter is to provide context on temporal variation in survival of first-year and adult little penguins over a 46-year period, as a pre-cursor to considering the climatic and oceanographic covariates that potentially explain some of the temporal variation. Included is a simple population model to investigate the stability of the population in relation to the observed trends.

4.1 Introduction

Long-term, continuous studies are necessary to investigate change in seabird de- mography and to determine the consequences of climate change [Clutton-Brock and Sheldon, 2010]. Reviews of studies which investigated links between temporal variation in population parameters of seabirds, such as breeding success and sur- vival, and climatic and oceanographic influences can be found in Grosbois et al. [2008], Croxall et al. [2002] and Sydeman et al. [2012]. The effects of climate on populations take multiple pathways and can be complex and lagged [Jenouvrier, 2013]. Using long-term data to investigate how survival varies with time is a useful pre-cursor to subsequent studies which attempt to link variation in survival and climatic and oceanographic variables.

Food supply is widely viewed as a significant factor in the regulation of population sizes of seabirds [Cairns, 1989]. Ashmole’s ‘halo effect’ model of population regula- tion proposed that large aggregations of animals may deplete food resources near the colony [Ashmole, 1963]. Furness and Birkhead [1984] postulated that seabird colony sizes are regulated by intercolony competition in common feeding grounds (the ‘hungry-horde’ model). An alternative model, the ‘hinterland’ model proposed by Cairns [1989], suggested that seabirds from neighbouring colonies occupy non- overlapping foraging areas, and the size of the colony is a function of the size of

57 the foraging area. The variation food supply over the length of study can provide useful insight into changes in seabird population.

On land, seabird populations are subject to predation, human impact [Halpern et al., 2008], and the direct effects of climate. Horswill et al. [2014] reported that the survival of macaroni penguins Eudyptes chrysolophus was negatively impacted by an increasing local population of the predator the giant petrel (genus Macronectes). Seven Antarctic seabird populations displayed varying responses to the interaction of human impact and environmental change: two species showed an increase and three species showed a recovery in their population after destruction of their habitat [Micol and Jouventin, 2001]. Bank cormorants Phalacrocorax neglectus nesting close to the water on Robben Island suffered nesting failures which were related to the direct effects of major storm events during the breeding seasons, suggesting that availability of suitable nesting sites may impose additional restrictions on population growth [Sherley et al., 2012b]. Thus, terrestrial variables can also provide reasons for variation in seabird populations.

To provide context for the rest of this thesis, I started by investigating the temporal variation in the survival of the little penguins used in my research. These penguins have been the subject of a long-running study that commenced in 1968 [Reilly and Cullen, 1979], with reports on many aspects of their biology; see for example Reilly and Cullen [1979, 1981, 1982, 1983], Dann et al. [1995], Johannesen et al. [2002b], Dann et al. [2005], Dann and Norman [2006], Ganendran et al. [2011], Sidhu et al. [2012] and Dann et al. [2014]. Previous work modelled age dependence in the survival, recapture and recovery probabilities simultaneously using mark- recapture-recovery data for this colony over a 36-year period (1968 − 2004) [Sidhu et al., 2007].

I used the little penguin dataset described in Chapter 2 to model the dependence of survival on time for each of two age classes, first-year and adult birds. A time- dependent model computes a survival estimate for each time period, here, each year. There are several important differences between my work and previous studies. Firstly, the identification of birds marked with either a flipper band or transponder has allowed separate survival estimates for banded and transpondered birds over the entire study period of 46 years. I did this by defining a covariate B which was set to be 1 for a banded bird and 0 for a transpondered bird (refer to Section 3.6 for model construction and notation). As there are no data for birds marked with both a flipper band and an injected transponder in the main dataset, I was not able to estimate the probability of tag loss, as in Dann et al. [2014]. Secondly, by including birds marked for the first time in only areas frequently visited for data collection, I attempted to ensure that a critical model assumption of homogeneous

58 recapture probabilities for all marked animals was satisfied (see page 32 for model assumptions). In reality the assumption of homogeneous recapture may be violated as some birds do not breed every year and may be less likely to be recaptured (refer to Section B.3 in Appendix B for a discussion on breeding data and recapture histories). Thirdly, and perhaps most significantly, I modelled survival of both age classes of birds with time. I fitted a series of models with different combinations of survival dependency which included both constant survival and time-varying survival. I used time as a covariate as part of the model-fitting process, rather than using a separate least-squares regression with the time-varying survival estimates as input data [Catchpole et al., 1999b]. Finally, I created a simple population model to investigate population stability in relation to observed trends.

4.2 Modelling time dependence in survival

4.2.1 The reference or full model

I first considered models in which each age class was kept constant, followed by models with time dependence for each age class in turn, and finally time depen- dence for both age classes (Models 4.1 – 4.4); in each successive model, different dependencies were introduced to see if the model provided a better fit. Models 4.5 – 4.8 included an indicator variable B that distinguished a flipper-banded bird from a transpondered bird (Section 3.6). All models had a recapture structure of p1, p2, p3 (age + time), as explained in Chapter 3. The results are shown in Table 4.1. (See Table 3.1 for an explanation of notation.)

Table 4.1: AIC values and number of parameters K for models involving time- dependent (t) and/or constant (C ) first-year (φ1) and adult (φ2) survival proba- bilities. Models 4.5 – 4.8 incorporated a banding effect for survival for both age classes. All models have a recapture structure of p1, p2, p3 (age + time).

Model K AIC

4.1 φ1(C), φ2(C) 49 1260.03

4.2 φ1(C), φ2(t) 93 651.03

4.3 φ1(t), φ2(C) 93 797.30

4.4 φ1(t), φ2(t) 137 211.14

4.5 φ1(B), φ2(B) 51 1106.03

4.6 φ1(t + B), φ2(B) 95 631.86

4.7 φ1(B), φ2(t + B) 95 480.68

4.8 φ1(t + B), φ2(t + B) 139 29.54

59 In Model 4.1, both first-year and adult survival were assumed constant. This model had the smallest number of parameters, and it was the worst-fitting model, with the highest AIC value. Models 4.2 and 4.3 introduced time variation in either first-year or adult survival while survival in the other age class remained constant. These models had a much larger number of parameters than Model 4.1; however the AIC values were much lower. Model 4.4, with both first-year and adult survival dependent on time had the largest number of parameters and a markedly lower AIC for this group of models.

The next group of models (Models 4.5 – 4.8) were similar to Models 4.1 – 4.4 except that they incorporated a banding effect. In Model 4.5, both first-year and adult survival were made dependent only on whether or not the bird was banded. Adding time dependence (in addition to banding effect) in turn to first-year survival (Model 4.6) and adult survival (Model 4.7) improved the model fit. In Model 4.8, survival in both age classes was assumed dependent on time and on whether or not the bird was banded. This model had the highest number of parameters and a dramatic improvement in AIC, confirming results in Dann et al. [2014] where a marked improvement in AIC was observed in survival models that included the effect of tagging. I have used Model 4.8 as the reference (or full) model in this thesis. The interactive effect of banding and time on survival is discussed in the next section.

The effects of banding are discussed in detail in Section 3.5.

4.2.2 Checking for overdispersion

Overdispersion (excess variation in the data) was discussed in Section 3.2.3; here I discuss the calculation ofc ˆ (the estimate of the variance inflation factor).

I used the annual survival and recapture estimates generated for the full model (Model 4.8, Table 4.1), and the number of birds in each cohort to simulate life- history data for the appropriate number of birds. For example, in 1980 there were 508 birds banded for the first time; I therefore used the survival and recapture estimates from 1980 to simulate life histories for each of those birds.

I simulated 99 datasets of life histories which were fitted to model Model 4.8 and the deviance was calculated each time. I then calculatedc ˆ (Section 3.2.3) to be 0.96. Since this value is very close to 1, I concluded that the data were not overdispersed and that the use of the AIC for model selection in my analysis was appropriate.

60 4.3 Variation in time-dependent survival

The transition from flipper banding to injected transponders on Phillip Island oc- curred from the late 1990s to the early 2000s (Table 2.8). The number of transpon- dered birds increased dramatically from a few hundred in 2001 to over a thousand in 2002, and occurred in response to observations by the researchers that banded birds were less likely to be recaptured. This decision was confirmed by findings of Dann et al. [2014] which showed that the survival of flipper-banded little penguins was significantly less than those with injected transponders. By 2004, almost all birds encountered for the first time were given transponders. Therefore, I used the time period 1969 – 2004 when analysing survival for banded birds and 2002 – 2012 for transpondered birds; this accounts for the different time scales in Figures 4.1 and 4.2. The dataset in my analysis did not contain birds marked with both tags as in Dann et al. [2014] so I was unable to estimate the probability of tag loss.

Figure 4.1 shows the dependence of survival on time (year) for banded birds from 1969 to 2004 using Model 4.8. The lines show the logit-linear relationships between survival and time, with time as a covariate in the survival models [Frederiksen et al., 2008]. I used Wald’s test to test for the significance of the slopes [Wald, 1943]. This test returns a p-value for the Wald test statistic, calculated as the square of the model regression coefficient divided by its standard error, and has a chi-squared distribution with v degrees of freedom (v= n − 1 where n is the number of observations). First-year survival showed a decreasing trend with time over the study period, with the slope of the line significantly different from zero (p < 10−2). The slope of the line for adult survival is not significantly different from zero (p=0.16), so adult survival remained essentially constant over the study period.

Figure 4.2 shows the dependence of survival on time (year) for transpondered birds from 2002 – 2011. As with banded birds, the slope of the line for transpondered first-year birds was significantly different from zero (p < 10−2), while that for adult survival was not (p = 0.08). The correlation between first-year and adult survival probabilities for banded birds is extremely low (r = 0.10, p = 0.58), suggesting that the causes of variation affecting first-year and adult survival for flipper-banded birds are different. This is also the case for transpondered birds, with a low correlation between first-year and adult survival (r =0.27, p=0.42).

The anomaly in Figure 4.2 for adult birds in 2012 may be an artefact of sampling, or an effect of parameter redundancy. Parameter redundancy can be present in models belonging to the Cormack-Jolly-Seber family of models (refer to Section

61 3.1) in which the last survival probability and the last recapture probability cannot be estimated separately, only their product being estimable [Gimenez et al., 2003].

1

0.9

0.8

0.7

0.6 adult

0.5

Survival 0.4

0.3 first−year 0.2

0.1

0 1970 1975 1980 1985 1990 1995 2000 2005 Year

Figure 4.1: Time-varying first-year and adult survival for banded birds. Vertical bars show standard errors; lines show the logit-linear relationship between survival and time, with time as a covariate in the survival model.

1

0.9

0.8

0.7 adult

0.6

0.5 Survival 0.4

0.3

0.2 first−year

0.1

0 2002 2004 2006 2008 2010 Year

Figure 4.2: Time-varying first-year and adult survival for transpondered birds. Ver- tical bars show standard errors; lines show the logit-linear relationship between survival and time, with time as a covariate in the survival model.

62 4.4 Population modelling

In long-lived organisms, population growth rate is more sensitive to variation in adult survival than to fecundity-related fitness components [Frederiksen et al., 2008]. A change in adult survival of little penguins produces a bigger change in population than a change in first-year survival or in breeding productivity [Dann, 1992]. In this section, I created a simple population model to investigate the stability of the population and to provide further support for the survival models used in my thesis.

The size of the population of little penguins on Phillip Island was estimated to be approximately 12,000 breeding penguins in 1985; this increased to around 35,000 breeding penguins in the late 1990s, decreasing slightly to 31,000 in 2010 [Sutherland and Dann, 2014]. Small-scale movements of birds into and out of a site may affect bird numbers, but this is difficult to estimate accurately because of a lack of data. Dann and Cullen [1990] suggested a movement of approximately 15% of individuals into and out of a study site; Sidhu [2007] estimated that the influx of breeders into a study site from the surrounding areas was about 12%. In this section I present a simple population model using a subset of breeding data, the data from one site, Location 29, Site 29, from 1986 – 2012 (details in Appendix B), from which I obtained estimates of fecundity and the movement of birds into and out of this site.

To create a simple population model, I needed to estimate the numbers of birds moving into the study site to breed. A useful indicator of this is the number of birds tagged for the first time as adults per cohort. I considered a breeding bird tagged for the first time at that site in a particular year to be a ‘stranger’, since a bird breeding at that particular site would have previously been recaptured. I was able to generate the total numbers of breeding birds seen for the first time each year, and calculated the proportion of ‘strangers’ at this site as 8.3%. Note that I used a subset of breeding data, the data from Location 29, Site 29, as this had the most breeding data. Mean values for chicks fledged per pair at this site showed good agreement with mean values calculated over all sites (correlation coefficient = 0.95, p < 10−2, Figure B.1, p. 143), indicating this site to be representative of the breeding population from all sites on Phillip Island.

Dann and Cullen [1990] have shown that approximately 50% of two-year-old birds are breeders, and that most birds are breeders from the age of three years. If I had only used two age classes for survival (for first-year and adult birds), as in the main analysis in this thesis, then I would have had to make an additional assumption about the proportion of birds in the second age class that were of breeding age. For this reason, I include in the Leslie matrix in the population model constant annual survival probabilities for the following three age classes of birds: first-year birds

63 (φ1), birds in their second year of life (φ2) and birds in their third and subsequent years of life (φ3), estimated using data from the length of the study and from all study sites considered in the main analysis. I used the same (age + time) structure for recapture probability as in all survival models in this thesis. Survival estimates were φ1 = 0.1258, φ2 = 0.7060 and φ3 = 0.7764; compare these with φ1 = 0.1638,

φ2 = 0.7174 and φ3 = 0.7896 reported in Sidhu et al. [2007].

The population model is written as Nt+1 = LNt where the vector Nt+1 is the number of birds in each age group in year (t + 1) and Nt is the distribution of ages at year t. L is the Leslie matrix

  0 φ2bF (φ3 + η)F   L =  φ1 0 0  , 0 φ2 φ3 + η where b is the proportion of breeding birds in their second year of life, F is a measure of fecundity (F = 0.54 chicks per individual) and η is the proportion of birds in their third and subsequent years of life that are immigrants from surrounding areas (η = 0.083) (data from Location 29, Site 29). I assumed that 50% of birds in their second year of life were breeding [Dann and Cullen, 1990] (b = 0.5), and that all birds in their third and subsequent years of life were able to breed. Note that I was not able to determine the number of birds per age group from the data, as the subset included adult birds of unknown age tagged for the first time.

These estimates gave a Leslie matrix with a principal eigenvalue λmax of approxi- mately 0.911, which predicts a population size that is very slowly decreasing. The slow decrease is in agreement with trends in estimates of penguin population size on Phillip Island [Sutherland and Dann, 2014], and importantly, provides confidence in the survival models used in my thesis.

4.5 Discussion

First-year and adult little penguins showed considerable time variation in annual survival over the 46 years. There was a statistically significant decrease in first- year survival for flipper-banded and transpondered birds, whereas there was no significant trend in the survival of adult flipper-banded or transpondered birds. In this section, I discuss some of the factors which may have accounted for the temporal variation in survival. This sets the context for the next three chapters in which I model survival with climatic and oceanographic covariates.

64 4.5.1 Survival and population numbers

The sample size of little penguins studied on Phillip Island increased over the length of the study as more burrows were added to the study area (Section 2.2). As men- tioned earlier, the population size of little penguins on Phillip Island was estimated at about 12,000 breeding penguins in 1985; increasing to around 35,000 breeding penguins in the late 1990s, decreasing slightly to 31,000 in 2010 [Sutherland and Dann, 2014].

The trends in the mean numbers of little penguins arriving ashore at Summerland Beach (‘Penguin Parade’) each night are a reasonable proxy for population trends over the colony [Sutherland and Dann, 2014]. The count of birds is made within a standard period (50 minutes from the arrival of the first bird) [Dann et al., 2000] by one observer with 10×40 binoculars, with assistance from secondary observers in conditions of poor visibility or when large numbers of penguins come ashore together [Sutherland and Dann, 2014]. Figure 4.3 shows the mean number of pen- guins crossing the beach at the Penguin Parade each evening from 1977 to 2012 [Sutherland and Dann, 2014] and, on the same graph, the first-year and adult sur- vival estimates from Model 4.8 to compare how observed population trends may be linked with the variation in penguin survival.

In the 1970s and 80s, colonies on Phillip Island came under immense anthropogenic pressures as well as predation (at its worst in the 1980s when there were sharp increases in penguin mortality on Phillip Island) [Dann, 1992]. Between 1986 and 1989, foxes Vulpes vulpes and dogs Canis familiaris accounted for approximately 70% of deaths [Dann, 1992]. Death by road accidents was a significant factor in areas of high human population, but decreased after 1984 when a traffic control system was installed [Dann, 1992]. Since 1985, there has been a significant pro- gram of habitat improvement which included the buy-back of residential areas, a reduction in weed species and erosion control [Dann, 1992]. Prior to 1986, first-year survival estimates were higher relative to subsequent years, while adult survival estimates were in general lower relative to subsequent years (Figure 4.3). During this period beach crossings were decreasing, indicating a relatively lower number of adult birds in the population. Therefore, the higher first-year survival was most likely a result of less competition with adults for food resources. As the breeding population increased relative to first-year birds, competition between the two age classes increased, and this was reflected in a decrease in first-year survival.

The number of beach crossings grew steadily from 1986 before showing a steep drop in 1995. This drop was reflected in the dramatic decrease in adult survival in 1995, followed by a decrease in first-year survival in 1996. In 1995 there was widespread

65 mortality of one of the main food species of little penguins, the sardine Sardinops sagax. This unprecedented die-off of sardines was reported as far west as Western Australia and halfway up the eastern coast of Australia, covering more than 5 000 km of coastline in the lower half of the continent. The cause was suggested to be a viral agent transmitted by migrating fish [Whittington et al., 1997]. The increase in penguin mortality in 1995 and decrease in breeding success in the following year were linked to this event [Dann et al., 2000]. The population has since remained largely stable; Sutherland and Dann [2014] suggest that this colony may be near its carrying capacity because there are no other colonies of little penguins larger than the Phillip Island colony. Nesting area for this colony does not appear to be a limiting factor on the population; Dann and Norman [2006] suggest that the population may level out with increasing intraspecific competition for food regardless of nesting area.

1 2000

adult

0.8

0.6

1000

0.4 Survival probability

beach crossings Average number of penguins 0.2 first−year

0 0 1970 1975 1980 1985 1990 1995 2000 2005 2010 2015 2020 Year

Figure 4.3: The mean number of penguins crossing the beach at the Penguin Parade each evening as a proxy for population size (data from Sutherland and Dann [2014]) and the time-varying survival for banded first-year and adult little penguins (error bars not shown).

In the period 1986 – 2004 there was a significant moderately strong positive correla- tion between beach crossings and adult survival of 0.57 (p = 0.01), indicating that an increase in beach crossings was associated with an increase in adult survival. There was also a significant moderately strong one-year lagged positive correlation of 0.63 (p < 10−2) between beach crossings and first-year survival. This indicated

66 that an increase in beach crossings in one year was followed by an increase in first- year survival in the following year.

4.5.2 Other factors affecting survival

Chicks are tagged just before fledging and most venture out to sea for the first time by the end of summer (austral summer, December to February). Around 80% of birds tagged as chicks are never seen again; most of their mortality occurs at sea [Reilly and Cullen, 1981, Dann, 1992, Sidhu et al., 2007]. Vulnerable and in- experienced first-year birds are more likely to succumb to parasitic infection and starvation [Harrigan, 1992], or may be more sensitive to variations in prey avail- ability [Agnew et al., 2016]. Laying date, banding date and banding weight have been found to be related to first-year survival of little penguins [Johannesen et al., 2003, Sidhu, 2007]. Chicks banded very early in the breeding season had greatly enhanced first-year survival probability (around 60%) compared with those banded later in the season (as low as 5%) [Sidhu, 2007], while those that fledged later had lower survival than those that fledged earlier [Chiaradia and Kerry, 1999]. Heavier chicks were also more likely to survive their first-year of life [Reilly and Cullen, 1982, Sidhu, 2007]. Other studies have linked hatching date and survival, for ex- ample, king penguin Aptenodytes patagonicus chicks on Possession Island, Crozet Archipelago, which hatched earlier in the season had a lower mortality rate than those which hatched later in the season, although this may have been confounded with experience [Stier et al., 2014]. In other seabird species, early hatched com- mon guillemot Uria aalge chicks on the Isle of May in Scotland were more likely to survive than chicks which hatched later in the season; this was attributed to ‘high-quality’ parents better able to prepare their chicks to survive their first winter at sea [Harris et al., 2007].

Adult little penguins are constrained to land during two energetically demanding events in their annual cycle: breeding and moult. They fast for a few days to a week during breeding and for a longer time (up to three weeks) during moult [Gales et al., 1988], making them subject to climatic and environmental effects while on land. In Chapter 5, I have investigated the survival of adult little penguins in relation to terrestrial climatic covariates such as air temperature, rainfall and humidity [Ganendran et al., 2016].

Oceanographic covariates and rough weather that could hamper foraging success would affect penguin survival at sea. For example, Sidhu et al. [2012] reported a positive association between first-year survival of flipper-banded and transpondered little penguins and sea-surface temperature in the waters in Bass Strait surrounding Phillip Island in autumn. A cooling of waters in northern Bass Strait in autumn

67 (March to May) has been associated with significantly later laying dates [Chambers et al., 2011b] and a reduction in first-year survival. Strong winds which caused per- turbations in the thermal stratification of coastal waters in Bass Strait, Australia, and a subsequent shift in prey availability were associated with reduced foraging efficiency in little penguins [Ropert-Coudert et al., 2009]. In Chapter 6, I have investigated the survival of little penguins in relation to wind and sea temperature.

Newly fledged little penguins disperse rapidly from Phillip Island to the west away from foraging waters used by adults closer to the colony [Collins et al., 1999]. In the peak of the breeding season in summer, it is very likely that food depletion occurs within close proximity to the colony where most of the adults birds feed [Dann and Norman, 2006]. This westward dispersion of newly fledged birds appears to be an adaptive strategy for inexperienced young birds facing high intraspecific competition with breeding adults for available food resources. In contrast, adult penguins remain close to the colony to provision offspring and maintain body reserves. In Chapter 7, I have investigated the survival of first-year and adult little penguins in relation to marine productivity.

The trade-off between reproduction and survival suggests that current reproduc- tive output should be negatively correlated with survival [Stearns, 1992]. Most adult mortality of the Phillip Island little penguins at sea was reported to occur from March to October which coincides with the period from the end of breed- ing/beginning of moult to the beginning of the next breeding season [Dann et al., 1992]. In a seventeen-month study of little penguins at Oamaru, New Zealand, Johannesen et al. [2002b] found that survival was lower during the post-moult pe- riod than at any other time of the year. A longer, five-year study of the same colony found a positive correlation between breeding success and survival over the breeding period [Johannesen et al., 2003]. However the authors cautioned that mortality during breeding could have been an artefact that influenced the positive correlation (that is a breeder surviving the breeding season would contribute to breeding success and vice versa) [Johannesen et al., 2003]. A positive relationship between breeding output per lifetime and longevity in little penguins on Phillip Island was reported by Dann and Cullen [1990]. In Appendix B, I have examined the relationship between breeding success and survival of little penguins on Phillip Island.

The confounding effects of age and experience on survival were reported in Nis- bet [2001] who showed that an apparent decrease in survival with age during the length of a study can be confounded with the increasing age of birds in that study. Although I have not been able to investigate the confounding effects of age since many of the birds in my analysis were tagged as adults of unknown age, Sidhu et al.

68 [2007] found a statistically significant decrease in survival of little penguins after the age of 9 years when both the survival and recapture parameters were allowed to vary with time. In contrast, no decrease in survival of older little penguins was found in the colony at Oamaru, New Zealand [Agnew et al., 2016]. A relationship between survival and age has been found in other seabird species, for example, the survival of adult great skuas Catharacta skua on the Shetland Islands in Scotland varied according to a quadratic model, with survival increasing in birds from 5 to 22 years old and decreasing sharply in birds over 22 years of age [Ratcliffe et al., 2002]. In a study of adult southern Buller’s albatrosses Thalassarche bulleri bulleriin New Zealand, Sagar et al. [2000] found evidence of lower survival of birds with at least 25 years of breeding experience compared with that of less experienced birds. Other studies have found strong negative relationships between adult survival and the number of years since first breeding, for example in northern fulmars Fulmarus glacialis [Dunnet and Ollason, 1978], short-tailed shearwaters Puffinus tenuirostr [Bradley et al., 1989] and Atlantic puffins Fratercula arctica [Harris et al., 1997].

Mention must be made here of the effects of banding on survival of little penguins in this colony as reported in Dann et al. [2014], who found that banded birds had lower survival rates than transpondered birds, and that the effect varied between years. Banding effect is discussed in detail in Section 3.5 and is included in the analysis in Chapters 5 – 7.

If adult survival of the little penguin population on Phillip Island remains constant as modelled here, it will, in the short-term, counteract the observed decreasing trend in first-year survival. Priority areas for research include the assessment and miti- gation of human recreational activities on penguin demography, and the modelling of effects of climate change on the penguin population [Dann, 2015].

In this chapter I have shown that the best model that describes time-variation in first-year and adult little penguins is one which is fully time varying and includes the effect of banding. A linear regression showed no statistically significant trend in adult survival and a decreasing trend in first-year survival. In the next three chapters, I investigate the extent to which the temporal variation in little penguin survival can be accounted for by climate and oceanographic covariates.

69

Chapter 5

Terrestrial climatic variables

The aim of this chapter is to investigate the seasonal effects of ter- restrial climatic covariates (ambient air temperature, rainfall and hu- midity) on adult little penguin survival in relation to their annual cycle, specifically during breeding and moult. This chapter is the basis of a manuscript entitled ‘Effects of ambient air temperature, humidity and rainfall on annual survival of adult little penguins Eudyptula minor in southeastern Australia’ [Ganendran et al., 2016].

5.1 Introduction

Seabirds are unique in that they use both marine and terrestrial habitats, and hence are exposed to the interaction of multiple climatic variables [Sydeman et al., 2012, Franklin et al., 2014]. Mitigation responses include shifts in geographic distribution between seasons, altered timing of natural events such as migration and breeding [Saino et al., 2007, Chambers et al., 2012, Sekercio˘glu et al., 2012], and behavioural adaptations [Frost et al., 1976, Oswald and Arnold, 2012].

Extreme weather can have significant effects on the survival of both chicks and adults. Strong winds and heavy rainfall during extreme storm events resulted in almost complete breeding failure in ivory gull Pagophila eburnea colonies in north- east Greenland [Yannic et al., 2014], while chick, and possibly egg, mortality of fork- tailed storm petrels Oceanodroma furcata breeding on the Barren Islands, Alaska, was linked to burrows which became wet in bad weather typified by heavy rainfall and high winds [Boersma et al., 1980]. Unprecedented rainfall in the Terra Ad´elie sector of Antarctica in the 2013/14 breeding season resulted in total chick mortality of Ad´eliepenguins Pygoscelis adeliae in that season [Ropert-Coudert et al., 2015]. Gales and heavy rainfall in the late winter were linked to large-scale mortality events for the European shag Phalacrocorax aristotelis on the Isle of May, Scotland [Frederiksen et al., 2008].

71 Seabird mortality can also be severely affected by extremely hot weather; for exam- ple, ambient temperatures of over 45◦C sustained for several hours can cause death by hyperthermia or evaporative water loss [McKechnie and Wolf, 2010]. A combi- nation of relatively high temperatures and the presence of mosquitoes feeding on the birds in high numbers was found to contribute to adult mortality of Br¨unnichs guillemot Uria lomvia, with mortality occurring among birds that were close to the edge of vegetation and exposed to high thermal loads in mid-afternoon [Gaston et al., 2002].

Penguins inhabit almost exclusively the temperate and polar regions of the south- ern hemisphere, and have evolved physiologically to adapt to cold conditions expe- rienced during extended periods at sea. Consequently, hot terrestrial habitats can challenge their ability to thermoregulate during periods of high ambient air temper- ature. The choice of nesting habitat of the African penguin Spheniscus demersus on Robben Island, South Africa, can affect breeding success, with artificial nests providing more protection from environmental factors such as heat stress than sur- face nests in the open [Sherley et al., 2012a]. Stokes and Boersma [1998] found that for Magellanic penguins, Spheniscus magellanicus, at Punta Tombo, Argentina, the use of nests with more cover was positively correlated with higher fledging success, possibly as a result of reduced exposure to predators and high temperatures. Tem- peratures inside artificial nest boxes on Penguin Island, Western Australia were found to be significantly higher than ambient air temperature, suggesting that high temperatures inside the boxes are a potential deterrent to adult occupancy, as well as contributing to chick mortality due to thermal stress [Ropert-Coudert et al., 2004a].

Little penguins are relatively intolerant to heat and cannot withstand prolonged exposure to air temperatures above 35◦C [Stahel and Nicol, 1988]. Rates of oxy- gen consumption increase at ambient temperatures above 25◦C [Baudinette et al., 1986]. Stahel and Nicol [1988] showed experimentally that little penguins exposed to an ambient air temperature of 35◦C became distressed and had a dramatically increased respiratory frequency. Body temperatures did not reach a steady state, but rose continuously throughout the heat exposure period. In Western Australia, prolonged exposure to high temperatures during the summer of 2008, with a higher- than-average number of days above 34◦C, may have contributed to increased mor- tality of little penguins moulting at this time [Cannell et al., 2011].

Unlike some seabirds which moult largely at sea away from breeding grounds, little penguins are constrained to land during moult. During this time, and during some breeding stages when they are also constrained to land, little penguins are subject to the direct effects of local climatic factors such as ambient air temperature, humidity

72 and rainfall. They fast for a few days to a week during breeding and for a longer time (up to three weeks) during moult, making these two life events demanding in terms of energy costs. Gales et al. [1988] found that the energy required to sustain a moulting little penguin was 15% higher than for a fasting, non-moulting little penguin resting on land. Little penguins lose weight during the breeding season, and put on 30 − 40% of their body weight in preparation for fasting during the moult. The post-breeding/pre-moult interval ranges from less than four weeks to six weeks, depending on when birds end their breeding [Reilly and Cullen, 1983].

Climate extremes can also occur in winter but I did not have any a priori reason to include climatic variables in winter as I was concerned with periods in the penguins’ life-cycle when they spent most of their time on land (breeding and moult). Chicks were marked at the age of approximately six to eight weeks and fledged at the age of eight to ten weeks. After fledging, chicks are rarely seen again on Phillip Island in their first year. Therefore it was not biologically sensible to consider the effects of terrestrial climatic variables on first-year survival.

Here, I have investigated the dependence of adult survival on ambient temperature, prolonged high temperatures, rainfall and humidity during the breeding and moult periods. To my knowledge, no previous studies has investigated how little penguin survival is impacted by terrestrial covariates during the periods when the birds are on land. I hypothesised that weather extremes or prolonged periods of high temperatures or rain influence adult survival negatively.

5.2 Data and methods

5.2.1 Penguin data

The penguin dataset used in this analysis was defined in Section 2.1.5.

5.2.2 Climate data

Data used to calculate maximum daily ambient air temperature and daily rainfall were provided by the Bureau of Meteorology, Melbourne, taken from the AWAP (Australian Water Availability Project; Jones et al. [2009]) gridded temperature data for the 0.05◦×0.05◦ grid points encompassing Summerland Peninsula (145◦90E, 38◦310S), for the period from 1968 to 2013. Humidity data were obtained from Laverton Royal Australian Air Force Base (37◦520S, 144◦460E, Figure 2.3, page 14), about 100 km northwest of the study area (see Section 2.6.1).

73 5.2.3 Model covariates

Using mean temperatures alone can mask the effects of extreme temperature events. In order to measure the cumulative effects of heat, I used number of degree-days, which is a proxy for prolonged high-temperature events. The concept of degree- days (or ‘growing degree-days’), which is widely used in agriculture and biology as a measure of heat accumulation [Ambrosini et al., 2011], originated in the 18th century [Bonhomme, 2000]. I adapted the calculation of degree-days to be the sum of all positive values of the difference between the daily maximum temperature and a threshold value, over the breeding or moult periods. Daily maximum temperature rather than daily mean temperature was used as this was a more accurate measure of thermal stress on little penguins; even a short period of time at temperatures above the threshold will affect penguins. I used threshold values of 27◦C and 35◦C. Baudinette et al. [1986] suggested that, at temperatures greater than 27◦C, the daily energy budgets of little penguins increased in tandem with increasing ambient temperatures as the birds started to hyperventilate in an attempt to reduce body temperature. At burrow temperatures higher than 35◦C, little penguins had only a few hours’ tolerance before body temperatures became dangerously high [Stahel and Gales, 1987].

Thus, for each of the breeding and moult periods I considered the effect of the follow- ing covariates on annual adult penguin survival: mean maximum daily temperature (‘tmp’); number of degree-days over 27◦C (‘degday27’); number of degree-days over 35◦C (‘degday35’); total rainfall (‘rn’); and mean daily humidity (‘hum’). I consid- ered the relationship between annual survival in calendar year j and the covariate in the breeding period from September in year j −1 to April in year j, and in the moulting period from February to April in year j [Chambers and Dann, 2015].

5.2.4 Methods

Methods, including model fitting and notation, and the formation of the likelihood, are described in detail in Chapter 3. Specifically, the interactive effects of the covariates and age on survival are described in Section 3.6.

5.3 Results

The mean covariate values (± standard error) are shown in Table 5.1.

Of all the climatic variables tested, the best-fitting first-stage covariate for adult survival was mean humidity during the moult period (hum mlt; Table 5.2). The hu- midity during the moult period over the length of the study ranged from a minimum of 12.1 hPa to a maximum of 15.3 hPa, with a mean value of 13.5 ± 0.1 hPa. There

74 Table 5.1: Mean covariate values (± standard error). The covariates were as follows: ‘tmp’ denotes the mean ambient temperature; ‘degday27’ the number of degree days over 27◦C; ‘degday35’ the number of degree days over 35◦C; ‘rn’ the total rainfall during the period; ‘hum’ the mean humidity during the period. The suffix ‘ brd’, indicates the breeding season from September to February and ‘ mlt’, the moulting season from February to April.

Covariate name Mean value ± standard error tmp brd 20.6 ± 0.1◦C tmp mlt 22.0 ± 0.1◦C degday27 brd 56.6 ± 4.1◦C degday27 mlt 16.7 ± 1.6◦C degday35 brd 3.6 ± 0.7◦C degday35 mlt 0.40 ± 0.01◦C rn brd 349.1 ± 13.1 mm rn mlt 155.7 ± 8.6 mm hum brd 12.3 ± 0.1 hPa hum mlt 13.5 ± 0.1 hPa was no statistically significant linear trend of this covariate with time (p = 0.07). The higher survival for the transpondered birds relative to the banded birds is also apparent. The positive regression coefficient (± standard error) of 0.182 (± 0.045) indicated that an increase in the mean humidity during moult was associated with an increase in adult survival.

Figure 5.1 displays the results of the best first-stage adult model for survival. There are two separate aspects to this graph: firstly, the time-dependent survival for banded birds (shown as stars) and transpondered birds (shown as open circles), and secondly, the models for covariate dependent survival (shown as a dashed line for banded birds and a solid line for transpondered birds).

Consider first the time-dependent survival: for each year there is a survival estimate (estimated using the full model, Model 4.8, Table 4.1). In addition, there is also a covariate value for each year. For example, in 1970 the estimates for adult survival were 0.77 for banded birds and 0.86 for transpondered birds. The value of the covariate in 1970 was 14.2 hPa. Therefore there is a star at the point (14.2, 0.77) and an open circle at the point (14.2, 0.86).

Consider next the models for covariate dependent survival: the dashed and solid lines are the models of predicted values for survival, logit (φ2,j) = b0 + b1X1,j + cB (Equation 3.3, adult birds) where X is the covariate, and B is the banded indicator variable. The dashed line is the model for banded birds and the solid line is the model for transpondered birds.

75 It is helpful to remember that the survival estimates shown on the graph are not in chronological order (for example, the survival estimates for 1970 are at the right of Figure 5.1), and are not data input to produce a ‘line of best fit’.

At the second stage, after allowing for the first-stage covariate, the best covariate was total rainfall during the moult period (rn mlt, Table 5.2). The negative re- gression coefficient (−0.002 ± 0.0003) indicated that an increase in total rainfall during moult was associated with decreasing adult survival. At the first stage of modelling, mean humidity during moult accounted for 11.7% of the deviance (time variation); the addition of the second-stage covariate increased the deviance ex- plained to 16.5%. The correlation between the first- and second-stage covariates was low, an absolute value of 0.15; at the second stage the regression coefficient (± standard error) of the best first-stage covariate was only slightly changed at 0.188 (± 0.021).

1

0.9

0.8

0.7 Adult survival

0.6

0.5

0.4 12 12.5 13 13.5 14 14.5 15 15.5 Mean humidity during moult period (hPa)

Figure 5.1: The relationship between adult survival under the fully time-varying model (Model 4.4) and mean humidity during the moult period for banded (stars) and transpondered (open circles) birds. The lines show the predicted values from the model for adult survival for banded (dashed line) and transpondered (solid line) birds for the fitted model of adult survival dependent on the mean humidity during moult.

At the third stage of modelling, after allowing for the first- and second-stage covariates, the best covariate was number of degree days over 35◦C during the breeding season (degday35 brd, Table 5.2). The negative regression coefficient (−0.033 ± 0.004) indicated that an increase in the number of degree days over

76 Table 5.2: Akaike’s information criterion (AIC) values for models which used each of the covariates listed for adult survival probability. The notation for the covariates appears in the caption to Table 5.1 All models used an (age + time) dependence for recapture probabilities, and first-year survival was fully time-varying (and depended on whether or not birds were banded, B). For each subsequent stage, the best model from the previous stage was used with an additional climatic covariate. The signs of the regression coefficients corresponding to each covariate are given. The regression coefficient (± standard error) as well as the proportion of time deviance explained by the best model at each stage are also given, in bold. The AIC of the null model was 631.86, and the AIC of the full model (Model 4.8), in which both first-year and adult survival probabilities were fully time-varying (and dependent on B), was 29.54. Stage 1 Stage 2 Stage 3 Stage 4 Covariate AIC Sign of reg. coeff. AIC Sign of reg. coeff. AIC Sign of reg. coeff. AIC Sign of reg. coeff. tmp brd 633.85 + 545.68 − 508.34 − 461.05 + tmp mlt 621.12 + 549.19 − 502.85 − 455.38 −

77 degday27 brd 633.77 + 553.42 − 515.06 − 456.28 + degday27 mlt 614.11 − 540.55 − 486.08 − 434.45 −0.007(±0.001) 29.8% degday35 brd 599.20 − 526.50 − 459.26 −0.033(±0.004) 25.9% degday35 mlt 633.85 + 547.16 + 523.08 + 452.11 + rn brd 609.63 + 538.45 + 491.60 + 457.52 + rn mlt 604.94 − 521.75 −0.002(±0.000) 16.5%

hum brd 596.13 + 553.56 + 514.48 + 458.68 + hum mlt 553.27 0.182(±0.045) 11.7% 35◦C during the breeding season was associated with decreasing adult survival. The addition of the third-stage covariate was well justified in terms of AIC (Table 5.2), and increased the deviance explained to 25.9%. The absolute values of the correlation coefficients for the third-stage covariate and the covariates of the first two stages were low (at around 0.03 for the first- and third-stage covariates and 0.09 for the second- and third-stage covariates).

At the fourth stage of modelling, the best covariate was number of degree days over 27◦C during the moult period (degday27 mlt, Table 5.2). The negative regression coefficient (−0.007 ± 0.001) indicated that an increase in this covariate was asso- ciated with decreased adult survival. The addition of this covariate to the model slightly increased the deviance explained to 29.8% (Table 5.2); the correlation coeffi- cients for this covariate and the first three covariates were low, with none exceeding 0.24.

None of the four covariates showed any significant linear trend over the length of the study period (with p-values for tests of significance of regression ranging from 0.07 to 0.56).

5.3.1 Interaction between banding and the best covariate for adult survival

Survival models used in this thesis included a banded indicator variable to determine the effect of banding on survival (Section 3.5). This is an additive effect such that the survival of banded and transpondered birds vary in parallel (Figure 5.1). In this section I also tested for an interactive banding effect to determine if the effects of the climatic covariate (from the best model) were different for banded and transpondered birds. Figure 5.2 shows the interactive effect of the mean humidity during the moult period and banding on adult survival. The positive regression coefficient for the survival of banded birds was statistically significant with this covariate (p < 10−2) and the negative regression coefficient for transpondered birds was not statistically significant (p = 0.9).

78 0.82

0.81

0.8 Transpondered birds

0.79

0.78

0.77 Adult survival 0.76 Banded birds

0.75

0.74

0.73

0.72 12 12.5 13 13.5 14 14.5 15 15.5 Mean humidity during moult period (hPa)

Figure 5.2: The interactive effect of banding and mean humidity during the moult period on adult survival.

5.4 Discussion

This study has provided the first evidence of the survival of little penguins being associated with terrestrial climatic variables during breeding and moult. I found that low humidity and high rainfall during the moult period had the largest negative impact on adult survival, followed by prolonged high temperatures during moult and breeding.

Humidity

Humidity had a strong positive correlation with survival in moulting penguins. Moult is a period of high energy cost for penguins, with a marked increase in energy expenditure required for feather synthesis and thermoregulation. Little penguins can lose approximately 46% of their body weight during this time [Gales et al., 1988]; the material lost includes fat and protein, but around half the loss is water [Croxall, 1982]. Williams et al. [1977] found that fasting, moulting Macaroni pen- guins Eudyptes chrysolophus and Rockhopper penguins Eudyptes chrysocome lose considerable quantities of water, also accounting for almost half the total weight lost during the moult fast. In the early stages of moult of emperor penguins Apten- odytes forsteri, new feather follicles covering the entire body are very rich in water which is lost during the final stage of moult [Groscolas, 1978].

79 The mitigating effects of high humidity in reducing water loss during the moult period may be the cause of the lower mortality of little penguins. It is puzzling that increased water loss may be the link between higher mortality and lower humidity, as penguins still have access to sea water at night to avoid dehydration. However, this is rarely observed (P Dann, pers. comm.).

Total rainfall

In this study, the total rainfall during the moult period had a negative association with adult survival. Little penguins moult after breeding, usually in the same area where successful breeding has occurred that season [Reilly and Cullen, 1983], but not always in the same burrow. They may moult in their burrows, in artificial nests or in more exposed areas, and moulting birds may also wander through the colony at night [Reilly and Cullen, 1983]. Moulting birds have been observed standing under open bushes for days at a time (L Renwick, pers. comm.). Artificial burrows are not waterproof and are subject to flooding, as are natural burrows, depending on the angle of the burrow and the substrate in which they lie. During wet conditions on Philip Island, some regularly checked burrows were damp or subject to minor flooding (L Chambers, pers. comm.). Burrows built on clay have a higher tendency to flood or become saturated with water than burrows in sand (L Renwick, pers. comm.). Moulting little penguins lose their old feathers in a short period of time, and new feather growth must be complete before they can return to sea [Stahel and Nicol, 1982]. During moult, a penguin’s plumage is not fully insulative, and birds would be more prone to hypothermia in wet conditions. These observations are consistent with findings here of a negative association between adult survival and mean total rainfall during moult.

Degree days with temperatures over 35◦C during breeding

The best third-stage covariate, number of degree days over 35◦C during the breed- ing period, had a negative association with adult penguin survival. Little penguins have a low ability to withstand heat stress [Stahel and Nicol, 1982]; a prolonged thermal load during breeding requires a substantial portion of the penguins’ energy budget to maintain body temperature. In periods of uncomfortably high temper- atures, penguins reduce body temperature by hyperventilating. Baudinette et al. [1986] have shown that evaporative water loss in captive moulting little penguins increased as the ambient temperature increased, and at temperatures up to 30◦C water loss for non-moulting birds increased exponentially. In the guard stage of chick rearing, if one parent is constrained to the nest because of a longer-than-usual foraging trip by the other parent, it is exposed to the increased physiological impacts

80 of the thermal load at the breeding site, causing it additional stress, potentially re- ducing its survival probability. It is well-documented that excessive heat can cause penguins to abandon breeding attempts in favour of survival (for example, Grif- fin [2005], Crawford et al. [2006] and Sherley et al. [2012a]). Heat can also cause mortality of juvenile and adult Magellanic penguins if they cannot find sufficient shade and cannot retreat to the water [Boersma and Rebstock, 2014]. Gal´apagos penguins Spheniscus mendiculus, while incubating eggs in the open and when faced with increasing thermal load from solar radiation, were unable to reduce their in- ternal body temperature by panting and abandoned the eggs to find relief in the ocean [Boersma, 1975]. Little penguins are strictly nocturnal when on land and will not retreat to the water to cool off during daylight hours unless they are extremely heat-stressed (P Dann, pers. obs.).

Degree days with temperatures over 27◦C during moult

The best fourth-stage covariate, number of degree days over 27◦C during the moult period, also had a negative association with adult penguin survival. This result is not surprising in light of the stressors faced by the birds at this time and their physiological response to heat as previously discussed. Daytime temperatures dur- ing the moult period of February to April can reach as high as 43.5◦C, with mean maximum temperature during moult over the length of the study of approximately 22◦C (Table 5.1, Figure 5.2). These temperatures were recorded using standardised data-gathering devices (most often a Stevenson Screen) in sheltered positions ele- vated at 1.2 m above ground level [Australian Bureau of Meteorology, 2011]. Actual temperatures at ground level in the sun can far exceed these, particularly on those hot days when there is little or no wind to mix the air [Australian Bureau of Me- teorology, 2007]. Moulting in shaded positions, for example, on south-facing slopes and under vegetation, is a likely adaptation to avoiding extremes of temperature.

Figure 5.3 shows the mean maximum temperature and mean total rainfall for each month over the study period, with a temperature peak in February.

Banding effects

The positive effect of humidity during moult on adult survival was particularly significant for banded birds, suggesting that any negative effects of banding may be offset by the positive effects on fitness of moulting adult birds of increasing humidity during the moult period. Although the model predicts a ‘crossover’ in survival of banded and transpondered birds, caution should be exercised when interpreting this effect as it occurs at the extreme end of the range of covariate values for which there are few data (Figure B.3, p. 150). The apparent slight drop in survival for transpondered birds is not statistically significant.

81 30 100

Mean total rainfall C) o

20 50 Mean total rainfall (mm) Mean maximum temperature Mean maximum temperature (

10 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month

Figure 5.3: Mean maximum temperature (◦C) (dashed line) and mean total rainfall (mm) (solid line) for each month over the length of the study period. Vertical bars represent one standard error either side of the mean.

5.5 Conclusions

In this study I have investigated the effects of some terrestrial climatic variables on survival of adult little penguins during breeding and moult on Phillip Island, south- eastern Australia. Humidity during moult had a positive association with adult annual survival, particularly for banded birds, possibly by alleviating water loss. Rainfall during moult had a negative association with adult survival. Prolonged heat during the breeding and moult periods also had a negative association with adult survival.

Breeding and moult are periods of high energy cost for little penguins. We have found that climatic conditions during the moult period, when birds are unable to go to sea, are more strongly associated with adult penguin survival than during the breeding period; birds abandon breeding if conditions on land become intoler- able [Frost et al., 1976]. Other studies have also found decreased survival of little penguins during the moult and post-moult periods [Johannesen et al., 2002b].

Decreases of up to 11% in regional annual rainfall are projected over the next few decades, but the intensity of heavy rainfall is expected to increase [Dann and Cham- bers, 2013]. Flooding and saturation of natural burrows and artificial nest boxes

82 may affect survival and breeding success. In addition, regional climate models pre- dict increases in ambient air temperature; the number of days with temperatures exceeding 35◦C is expected to double by 2030 and triple by 2070, with accom- panying increases in the incidence of drought and heat waves [Chambers et al., 2011b, Chambers and Dann, 2015]. Apart from the effects on penguin mortality of hyperthermia and heat stress caused by higher temperatures, the increase in temperatures together with decreasing rainfall increase the risk of fire. Substan- tial changes in vegetation as a result of fire is thought to be a contributor to the loss of a little penguin colony in Tasmania [Stevenson and Woehler, 2007]. Fires have historically caused death and injury to little penguins on Phillip Island, with several major fires occurring since the 1960s. Little penguins have been reported to have inappropriate and maladaptive responses to fire, with many remaining in their burrows where they died and others standing beside flames preening singed feathers instead of moving away [Chambers et al., 2011a].

Although the Phillip Island colony of little penguins appears secure, with terrestrial predators and anthropogenic threats largely eliminated and sufficient area available for breeding, longer-term climate changes on land and at sea will influence popula- tion size [Sutherland and Dann, 2014]. The development of appropriate adaptation actions, such as improved buffering of extreme weather through artificial-burrow designs and increased complexity of vegetation structure in breeding areas (see also Dann and Chambers [2013] and Hobday et al. [2014]), will further contribute to the conservation of seabirds which breed and moult on land.

83

Chapter 6

Wind strength and sea temperature

The aim of this chapter is to investigate the effects of seasonal cli- matic and oceanographic covariates which operate at sea (wind and sea temperature) on first-year and adult little penguin survival in relation to their annual cycle. The results from this chapter form the basis of a manuscript enti- tled ‘The effects of local wind strength and ocean temperature on the survival of little penguins Eudyptula minor in southeastern Australia’ [Ganendran et al,. in prep.].

6.1 Introduction

It is well established that climate plays a critical role in the ecology of seabirds, af- fecting them directly, both physiologically and behaviourally, and indirectly through their food chains and competitors [Ainley et al., 2010, Barbraud et al., 2010]. Cli- matic influences such as wind and sea temperature may affect timing of life-history events and movement or distribution of seabirds [Jones et al., 2002, Crespin et al., 2006, Weimerskirch et al., 2012] and their prey [Becker et al., 2007, McLeod et al., 2012, Montero-Serra et al., 2015].

As upper-trophic marine predators, seabirds are sensitive to long-term climate trends, as well as to shorter-term environmental factors [Reid and Croxall, 2001, Croxall et al., 2002, Weimerskirch et al., 2003]. Recent climate changes have been linked to shifts in the demography of seabirds, affecting species abundance, distribu- tion and phenology [Sydeman et al., 2001, Crick, 2004, Sandvik et al., 2005, Hipfner, 2008]. Phenomena measured by climate indices, such as the El Ni˜noSouthern Oscil- lation and North Atlantic Oscillation, have complex influences on widely dispersed seabird species [Thompson and Grosbois, 2002, Jenouvrier et al., 2005b, Hemery et al., 2008, Sandvik et al., 2008, Ancona et al., 2012], and are linked to long- term changes in seabird population dynamics. Determining plausible mechanisms that explain biological responses to climatic influences is complex but can provide

85 useful associations between seabird population parameters and climatic variables [Congdon et al., 2007, Forcada and Trathan, 2009].

Ocean warming, a consequence of climate change, is likely to have a profound in- fluence on pelagic ecosystems and accordingly, the demography of seabirds [Hodder and Graybill, 1985, Hyrenbach and Veit, 2003, Inchausti et al., 2003, Breton et al., 2008, Chambers et al., 2011b]. These influences can be lagged, and can precede the resulting life-history events by months or years [Mickelson et al., 1992, Decker et al., 1995, Durant et al., 2003, Sidhu et al., 2012], suggesting mechanisms operating at a trophic level which cause a temporal or spatial mismatch of seabirds and prey. Effects can also be physical, such as the presence of thermoclines presenting barriers to movement of nutrients or prey, which could affect the access of marine birds to the prey [Bost and Le Maho, 1993, Gr´emilletet al., 2009, Ropert-Coudert et al., 2009]. Sea-surface temperature (SST) has been shown to have a lagged effect on the survival of adult Atlantic puffins Fratercula artica [Harris et al., 2005]. Warm-water episodes in the winter before breeding have been found to have a lagged effect on body condition, which resulted subsequently in depressed breeding performance of the blue petrel Halobaena carulea on the Kerguelen Islands [Guinet et al., 1998]. Higher SST in the pre-breeding season was linked with reduced breeding success of little penguins on Penguin Island, Western Australia [Cannell et al., 2012]. In contrast, Chambers [2004a] and Cullen et al. [2009] found that a high local SST at Phillip Island, particularly in autumn, was associated with an early start to breeding, leading to greater breeding success. Sidhu et al. [2012] found a negative association between first-year survival and an east-west sea temperature gradient in Bass Strait in winter, and a positive association with the mean SST temperature in autumn.

Wind strength might also affect seabird population dynamics, possibly via increased energy costs in response to wind-induced ocean conditions, or via disruption in prey availability [Saraux et al., 2016]. Studies on the effects of wind on seabirds have focussed on the flight behaviour of and energetic costs to seabirds (Spear and Ain- ley [1997], Weimerskirch et al. [2000], Felic´ısimoet al. [2008], Suryan et al. [2008], Adams and Flora [2010]), while others have linked oceanic winds to foraging pat- terns and prey availability [Schroeder et al., 2009, Raymond et al., 2010, Lewis et al., 2015]. Wind patterns would not have the same effect on travel time of penguins as they would on flying seabirds, instead affecting penguins via prey dispersion and availability [Ropert-Coudert et al., 2009, Dehnhard et al., 2013] or by rough seas hampering foraging efforts [Berlincourt and Arnould, 2015]. Other studies have found links between winds from particular directions and penguin breeding success

86 [Mickelson et al., 1992]. Previous work by Ganendran et al. [2011] found that first- year survival of little penguins in southeastern Australia was positively associated with mean southerly wind speed in the winter prior to a chick’s hatching, and that adult survival was positively associated with mean northerly wind speed in the autumn following moult. The same study also found that the survival of both first- year and adult birds was negatively associated with mean easterly summer wind speed.

Given the potential for wind and SST patterns to affect seabird demography, the aims of this study were to examine whether and how little penguin survival has been impacted by these variables. A westerly wind blowing across Bass Strait was previously found to have a positive association with penguin body weight [Mickelson et al., 1992]. Therefore I wished to investigate further whether a westerly wind was associated with penguin survival. I also investigated how increasing wind speed, strong wind events and SST potentially affected penguin survival. Building on previous work which found an association between SST and first-year survival [Sidhu et al., 2012], I examined the association between SST and the survival of adult birds. Furthermore I considered models which included both wind and SST covariates to see if they had more predictive power for little penguin survival than models with either covariate alone. To my knowledge, no such studies linking a combination of these covariates and little penguin survival have been previously reported.

6.2 Data and methods

6.2.1 Penguin data

The penguin dataset used in this analysis was defined in Section 2.3.5.

6.2.2 Climatic data

Wind data were obtained from the official database of the Australian Bureau of Meteorology, station number 87031, Laverton RAAF Base (37◦520S, 144◦460E, ele- vation 16 m, Figure 2.3, page 14), about 100 km northwest of the study area, and just northwest of Port Phillip Bay. This site was used for several reasons, primarily because it was the closest site with reliable wind data over the entire study period, and also for consistency with previous studies [Mickelson et al., 1992]. The use of this site was further justified by the fact that adult penguins spend a considerable amount of time in Port Phillip Bay in winter and early spring [McCutcheon et al., 2011]. Wind data comprised twice-daily (9:00 am and 3:00 pm) readings of wind speed (m/s) and direction.

87 SST data were based on the U.S. National Centers for Environmental Prediction (NCEP) data from 1949 – 2001 (after which time the format changed) and corrected satellite data from 2001 – 2013 [Reynolds and Smith, 1994]. The edges of the grid of the SST data run from 138◦E to 152◦E and from 35◦S to 45◦S, with a 1◦ × 1◦ resolution. For each month there are 10 rows and 14 columns of data, with each data point representing the monthly mean SST at the centre of each grid square. Cells with centres on land were ignored. Further explanation of the SST data can be found in Chambers [2004b].

6.2.3 Model covariates

Mean wind speed

The mean wind speed was calculated by averaging the two daily wind-speed read- ings, then calculating the mean values for the period. Wind direction was not taken into account in the calculation of this covariate as my aim here was to determine whether the overall mean wind speed could be used to predict penguin survival.

East-west wind speed

To calculate a daily east-west wind-speed component, I first calculated the east- west (u) component for each 9:00 am and 3:00 pm reading using the equation u = speed × −sin(direction), then calculated the daily mean of those two values. To obtain the mean for a period, I used the daily means for the relevant months; for example, to obtain the autumn mean (austral autumn: March, April and May), I averaged the daily means for those three months. A positive value indicated wind from the west, a negative value wind from the east. This variable provided a measure of the nett effect of easterly and westerly winds, and its use as a covariate for survival offered some insight into the hypothesis that westerly winds accelerate the flow of cool, nutrient-rich waters into Bass Strait, or conversely, that strong easterly winds impede the westerly flow of nutrients.

For standard averaging procedures for wind measurements, see for example www.ndbc.noaa.gov/wndav.shtml or www.webmet.com/met monitoring/622.html.

Strong-wind events

Regional patterns of wind are predicted to shift poleward and increase in intensity due to climate change. Hence I included in this analysis a covariate that is a measure of strong-wind events. I first defined a day to be of ‘strong wind’ if it had at least one wind speed reading for the day that was greater than the 90th percentile value of 32.4 km/h (9.0 m/s, calculated from the maximum daily values). I refer to these

88 days as ‘strong-wind days’. To create the covariate representing repeated strong- wind events, I then calculated the number of strong-wind days for each season. It is interesting to note that wind speeds greater than 30 km/h represent fresh winds on the Beaufort Wind Scale and are associated with moderate waves and crested wavelets on inland waters (see www.bom.gov.au/lam/glossary/beaufort.shtml).

Sea-surface temperature (SST) and sea temperature gradient

I used the same SST study area as in Sidhu et al. [2012] (Figure 6.1) for consistency. SST was averaged over the region 143 – 145◦E, 38 – 40◦S (labelled ‘SST Box’, Figure 6.1). The two sites used in the calculation of the east-west sea temperature gradient are also shown. Site 1 covered 145 – 146◦E, 39 – 40◦S; Site 2 149 – 150◦E, 38 – 39◦S. The east-west sea temperature gradient was defined as the difference in SST between Site 2 and Site 1. A positive value indicated a higher SST at Site 2 than Site 1, a negative value a lower SST at Site 2 than Site 1.

Figure 6.1: Map showing the locations of Phillip Island (38◦310S, 145◦100E), the sea-surface temperature (SST) box and the two sites used to measure the sea tem- perature gradient in Bass Strait (Site 1 and Site 2). The SST box covers 143−145◦E, 38−40◦S; Site 1 covers 145−146◦E, 39−40◦S; Site 2 covers 149−150◦E, 38−39◦S. Also shown is a schematic representation of the currents in the Bass Strait region. Summer currents are denoted by dashed lines. EAC: East Australian Current; SAC: South Australian Current; ZC: Zeehan Current; FC: Flinders Current; CC: Summer Coastal Current; BSC: Bass Strait Cascade; SAW: Subantarctic Water. Adapted from Sandery [2007].

89 Table 6.1: Wind and SST covariates.

Covariate label Description WS ‘Mean wind speed’; for example ‘WS WiP’ was the mean wind speed in the previous winter. WD ‘Strong-wind days’; for example ‘WD WiP’ was the number of days in the previous winter with at least one wind-speed reading for the day greater than the 90th percentile value of 32.4 km/h. I referred to these as ‘strong-wind days’. EWW ‘East-west wind’; for example ‘EWW WiP’ was the mean east-west wind speed in the previous winter. A positive value indicates wind from the west, a negative value wind from the east. SST ‘Sea-surface temperature’; for example ‘SST WiP’ was the mean sea-surface temperature in the previous winter. TGR ‘Temperature gradient’; for example ‘TGR WiP’ was the mean east-west sea temperature gradient between Site 2 and Site 1 in the previous winter, calculated as the difference in SST between the two sites. A positive value indicates a higher SST at Site 2, a negative value a higher SST at Site 1. Suffixes Description WiP Previous winter SpP Previous spring SuC Current summer AuC Current autumn WiC Current winter SpC Current spring

When considering the effects of climatic and oceanographic influences on penguin survival, seasonal rather than monthly covariates were used to decrease the number of covariates and so reduce the chances of producing spurious correlations [Federer, 1955, Burnham and Anderson, 2002]. In addition, seasonal rather than monthly values were more appropriate, as the phenomena which cause mortality for little penguins tend to span months, but are less likely to span seasons [Dann et al., 1992].

To account for lagged effects, I allowed survival over calendar year j to depend on covariates over six seasons (an 18-month period), from the Winter of the Previous calendar year (1 June – 31 August in calendar year j−1) denoted by ‘WiP’, to the Spring of the Current year (1 September – 30 November in calendar year j ) denoted by ‘SpC’. See Table 6.1 for a full list of covariate labels and their description.

90 6.2.4 Methods

Methods, including model fitting and notation, and the formation of the likelihood, are described in detail in Chapter 3. Specifically, the interactive effects of the covariates and age on survival are described in Section 3.6.

6.3 Results

Tables 6.2 and 6.3 contain values of the Akaike’s information criterion (AIC) for models using seasonal wind and sea-temperature covariates for first-year and adult survival probability, respectively. The AIC of the null model for first-year survival was 480.68, while the AIC of the null model for the adult survival was 631.86. The null model for first-year survival has first-year survival depending only on whether or not a bird was banded (covariate B) and adult survival fully time-varying and depending on whether or not a bird was banded; similarly for the null model for adult survival. The AIC of the full model, in which both first-year and adult survival probabilities were fully time varying (and dependent on B) was 29.54 (see also Table 4.1 on page 59).

6.3.1 First-year survival

First stage

The best-fitting first-stage covariate for first-year survival was mean east-west sea temperature gradient in the previous winter (TGR WiP, AIC = 399.15, Table 6.2), with the negative regression coefficient indicating that an increase in the mean east- west sea temperature gradient across Bass Strait in the winter of the previous year was associated with a decrease in first-year survival in the current year.

However, this covariate showed an increasing linear (temporal) trend over the study period (p < 10−5, Figure 6.2); thus the apparent relationship between the covariate and first-year survival may have been due to the temporal trend, which could mask the true effects of the covariate. Grosbois et al. [2008] suggested that climatic covariates which exhibit linear trends over time may result in spurious relationships with species survival, since it would be difficult to rule out the possibility that any trend in survival resulted from a relationship with another causal factor that also exhibited a trend, rather than from a causal relationship with the climatic covariate of interest. Here, the observed negative association between first-year survival and the sea temperature gradient in the previous winter may have been due to first-year survival decreasing and mean east-west sea temperature gradient increasing over time (see Chapter 4 for discussion on the temporal variation of first-year survival).

91 Table 6.2: Akaike’s information criterion (AIC) values for models using the covariates listed for first-year survival probability. All models used an (age + time) dependence for recapture proba- bilities, and adult survival was fully time-varying (and depended on whether or not birds were banded, B). For each subsequent stage, the best model from the previous stage was used with an additional climatic covariate. The signs of the regression coefficients corresponding to each covariate are given. The regression coefficient (± standard error) as well as the proportion of time deviance explained by the best model at each stage are also given, in bold. Results for Stage 4 models are not shown due to their large number. The suffix ‘ d’ denotes de-trended covariates (residuals).

Stage 1 Stage 2 Stage 3 Covariate AIC Sign of reg. coeff. AIC Sign of reg. coeff. AIC Sign of reg. coeff. EWW WiP 438.22 + 395.29 + 378.36 + EWW SpP 414.97 0.325 (±0.060) 12.6% EWW SuC 479.61 − 415.50 + 387.10 + EWW AuC 452.77 − 394.90 − 363.88 − EWW WiC 480.10 + 408.00 + 388.47 − EWW SpC 476.53 + 399.49 + 376.85 + WS WiP 474.57 + 414.49 + 386.88 + WS SpP 478.31 + 387.30 − 376.54 − WS SuC 482.68 − 416.91 − 388.92 − WS AuC 476.19 − 416.47 − 387.37 − WS WiC 451.91 + 402.08 + 384.90 − WS SpC 478.22 − 415.67 − 382.86 − WD WiP 427.32 + 371.37 + 352.66 + WD SpP 469.07 + 410.50 − 388.39 − WD SuC 482.54 + 416.88 + 387.85 + WD AuC 475.71 + 395.95 + 373.87 + WD WiC 409.31 + 360.67 + 336.23 + WD SpC 461.05 − 408.52 − 374.37 − SST WiP 448.27 + 392.59 + 364.46 0.345 (±0.063) 22.7% SST SpP 479.89 + 397.90 + 363.66 + SST SuC 482.47 − 416.86 + 388.44 + SST AuC 477.33 + 404.10 + 375.20 + SST WiC 482.67 − 415.50 + 388.24 + SST SpC 476.65 − 413.62 − 386.22 − TGR WiP 399.15 − 401.15 − 337.62 − TGR SpP 475.04 − 387.76 + 382.04 − TGR SuC 481.99 + 387.24 + 388.93 + TGR AuC 480.17 − 401.04 + 372.17 − TGR WiC 478.05 − 401.04 − 379.68 − TGR SpC 480.53 − 400.05 + 384.49 − TGR WiP d 431.21 − 401.15 − 366.94 − WD WiC d 436.40 + 386.95 0.026 (±0.005) 18.1% WD WiP d 391.13 + 370.18 +

92 2.6 C) o

2.4

2.2

2

1.8

1.6

1.4

1.2

1

0.8 Mean east−west sea temperature gradient in previous winter ( 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year

Figure 6.2: Mean east-west sea temperature gradient in the previous winter (with trend line added) between 1967 and 2011.

0.8

0.6

0.4

0.2

0

−0.2

−0.4

−0.6 Residuals of east−west sea temperature gradient in previous winter

−0.8 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year

Figure 6.3: Residuals of mean east-west sea temperature gradient in the previous winter (with trend line added) between 1967 and 2011.

93 Grosbois et al. [2008] suggested the removal of the linear trend (‘de-trending’) by taking the residuals of the covariate (referred to as the ‘de-trended covariate’) and using the de-trended covariate in the model. A residual is the difference between an observed value and the predicted value (obtained from a linear regression model of the climatic covariate over time). Here, using the de-trended covariate resulted in an increase in the AIC (TGR WiP d, AIC = 431.21, Table 6.2 and Figure 6.3), meaning that it was no longer the best covariate in the first stage of modelling. I therefore concluded that the effect of this covariate (TGR WiP) may have been confounded with time, and excluded it from the model.

Once TGR WiP was excluded, the next-best first-stage covariate for first-year sur- vival was number of strong-wind days in the current winter (WD WiC), with the positive regression coefficient indicating that an increase in the number of strong- wind days in the current winter was associated with an increase in first-year sur- vival. However, this covariate also showed a linear trend (p < 10−3, Figure 6.4), suggesting the possibility of a spurious result. This time both first-year survival and the covariate were decreasing over time, perhaps resulting in the observed pos- itive association. As before, the de-trended covariate was used in the modelling (Figure 6.5 for the de-trended covariate). Again, this resulted in an increase in AIC (AIC = 436.40, Table 6.2). I therefore also excluded this covariate.

35

30

25

20

15

Number of strong wind days in current winter 10

5 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year

Figure 6.4: Number of strong-wind days in the current winter (with trend line added) between 1968 and 2012.

94 15

10

5

0

−5

−10 Residuals of the number strong wind days in current winter

−15 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year

Figure 6.5: Residuals of number of strong wind days in the current winter (with trend line added) between 1968 and 2012.

The next-best first-stage covariate for first-year survival was mean east-west wind speed in the previous spring (EWW SpP), with the positive regression coefficient (± standard error) of 0.325 (± 0.060) (AIC = 414.97, Table 6.2) indicating that an increase in the mean east-west wind speed in the previous spring was associated with an increase in first-year survival. This covariate did not show any linear trend over the study period (p = 0.78, Figure 6.6). At the first stage this covariate accounted for 12.6% of the deviance (time variation).

95 4

3.5

3

2.5

2

1.5

1

Mean east−west wind speed in previous spring (m/s) 0.5

0 1965 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year

Figure 6.6: Mean east-west wind speed in previous spring (with trend line added) between 1967 and 2011.

0.6

0.5

0.4

0.3 First−year survival 0.2

0.1

0 0 0.5 1 1.5 2 2.5 3 3.5 4 East−west wind in previous spring (m/s)

Figure 6.7: The relationship between time-dependent first-year survival under model φ1(t+B) and mean east-west wind speed in the previous spring for banded (stars) and unbanded (open circles) birds. The lines show the predicted values for first-year survival for banded (dotted line) and transpondered (solid line) birds for the fitted model of first-year survival dependent on mean east-west wind speed in the previous spring. 96 Second stage

Once mean east-west wind speed in the previous spring was accounted for, the best- fitting covariate at the second stage was number of strong-wind days in the current winter (WD WiC), with the positive regression coefficient (± standard error) of 0.034 (± 0.004) indicating that an increase in the number of strong-wind days in the current winter was associated with an increase in first-year survival. However, as mentioned previously, this covariate showed a linear trend (p < 10−3, Figure 6.4). Using the de-trended covariate resulted in a model with an AIC value of 386.95 (WD WiC d, Table 6.2), which is higher than several other models, and I therefore excluded this covariate.

The next-best covariate was number of strong-wind days in the previous winter (WD WiP, Table 6.2), with the positive regression coefficient indicating that an in- crease in the number of strong-wind days in the previous winter was associated with an increase in first-year survival in that year. However, the linear trend displayed by this covariate (p < 10−3, Figure 6.8) led me to use the de-trended covariate, which resulted in a model with an increased AIC (WD WiP d, AIC = 391.13, Table 6.2) that was higher than some other candidate models. I therefore excluded this covariate.

Once WD WiP d was excluded as a possible best second-stage covariate, the next- best covariate was in fact WD WiC d, the de-trended (residuals of) number of strong-wind days in the current winter, with a positive regression coefficient. In the context of this analysis, an increase in WD WiC d was an indication that the number of strong-wind days was increasing relative to its mean value. Thus, the observed positive association meant that an increase in first-year survival was asso- ciated with a higher-than-average number of strong-wind days in the current winter. The correlation between the first-stage covariate (EWW SpP) and the de-trended covariate (WD WiC d) was low, with an absolute value of the correlation coefficient of 0.10. The regression coefficient (± standard error) of the first-stage covariate was only slightly changed, at 0.295 (± 0.041), from its value at the first stage. The addi- tion of this covariate increased the proportion of deviance (time variation) explained by the model from 12.6% to 18.1%.

97 35

30

25

20

15

Number of strong wind days in previous winter 10

5 1970 1975 1980 1985 1990 1995 2000 2005 2010 Year

Figure 6.8: Number of strong-wind days in the previous winter (with trend line added) from 1967 to 2011.

20

15

10

5

0

−5

−10 Residuals of number strong wind days in previous winter

−15 1970 1975 1980 1895 1990 1995 2000 2005 2010 Year

Figure 6.9: Residuals of number of strong-wind days in the previous winter (with trend line added) from 1967 to 2011.

98 There were three other second-stage models with virtually indistinguishable AIC values; the difference in AIC values between the second-stage model containing WD WiC d and each of these three models was less than one unit. These three covariates bear mentioning here: mean east-west sea temperature gradient in the current summer (TGR SuC, AIC = 387.24); mean wind speed in the previous spring (WS SpP, AIC = 387.30); and mean east-west sea temperature gradient in the pre- vious spring (TGR SpP, AIC = 387.76, Table 6.2). Of these three covariates, only WS SpP showed no linear trend (p = 0.10), while the other two showed strong linear trends (p < 10−3 for both). However, the correlation between the best first-stage covariate for first-year survival (EWW SpP) and WS SpP was moderately strong (correlation coefficient 0.66). Therefore, it would be unwise to include any of these three covariates in the model since it would not be possible to distinguish the unique effects on first-year survival of each covariate. Therefore, despite relatively low AIC values, models containing these three covariates were not considered.

Third and fourth stages

Again, a covariate was excluded if it displayed a linear trend and the de-trended covariate showed an increase in AIC or was correlated to the best-fitting covariates of earlier stages.

The two best-fitting covariates at the third stage, number of strong-wind days in the current winter (WD WiC) and mean east-west sea temperature gradient in the previous winter (TGR WiP, Table 6.2) showed a linear trend as discussed earlier, and use of the de-trended covariates showed increases in AIC values. Therefore both of these covariates were excluded.

There were three other third-stage models with AIC values all within one unit of each other: mean east-west wind speed in the current autumn (EWW AuC); sea- surface temperature in the previous winter (SST WiP); and sea-surface temperature in the previous spring (SST SpP, Table 6.2). None of these covariates displayed a linear trend (p = 0.69, 0.68 and 0.81 respectively). The mean east-west wind speed had a negative association with first-year survival, while the two sea-surface temperature covariates had positive associations. Each of these covariates accounted for approximately 23% of the proportion of deviance (time variation).

Stage 4 models were run using each of these three covariates as the best third- stage covariate in turn (results not shown due to the large number of models run, computer code available from [email protected]). Many of the models had to be excluded because the covariates displayed either a linear trend (with increases in AIC when the de-trended covariates were used) or were correlated with earlier stage covariates. After these exclusions, the best third-stage covariate

99 was mean sea-surface temperature in the previous winter (SST WiP, AIC = 364.46), with a positive association with first-year survival (regression coefficient (± standard error) of 0.345 (± 0.063). The third-stage covariate increased the proportion of deviance (time variation) explained by 4.6% to 22.7%.

The best fourth-stage covariate (with AIC = 345.8) was mean east-west wind speed in the current autumn (EWW AuC), with a negative association with first-year sur- vival regression coefficient (± standard error) of −0.255 (± 0.049)
. This covariate had no linear trend, and only weak correlations with the first-, second- and third- stage covariates. The fourth-stage covariate accounted for a further 3.3% of the deviance, bringing the total proportion of deviance explained by the fourth-stage model to 26.5%.

6.3.2 Interaction between banding and the best covariate for first-year survival

Figure 6.10 shows the interactive effect of the mean east-west wind speed in the previous spring and banding on first-year survival. The negative regression co- efficient for the survival of banded birds was not statistically significant for this covariate (p = 0.9) and the positive regression coefficient for transpondered birds was statistically significant (p < 10−2).

0.45

0.4

0.35

transpondered birds 0.3

0.25

0.2

First−year survival 0.15

0.1 banded birds 0.05

0 0 0.5 1 1.5 2 2.5 3 3.5 4 East−west wind in previous spring (m/s)

Figure 6.10: The interactive effect of banding and mean east-west wind speed in the previous spring on first-year survival.

100 6.3.3 Adult survival

Table 6.3: Akaike’s information criterion (AIC) values for models using the covari- ates listed for adult survival probability. All models used an (age + time) depen- dence for recapture probabilities, and first-year survival was fully time-varying (and depended on whether or not birds were banded, B). For each subsequent stage, the best model from the previous stage was used with an additional climatic covariate. The signs of the regression coefficients corresponding to each covariate are given. The regression coefficient (± standard error) as well as the proportion of time de- viance explained by the best model at each stage are also given, in bold. The suffix ‘ d’ denotes de-trended covariates (residuals).

Stage 1 Stage 2 Stage 3 Stage 4 Covariate AIC Sign of reg. coeff. AIC Sign of reg. coeff. AIC Sign of reg. coeff. AIC Sign of reg. coeff. EWW WiP 633.83 − 527.71 + 478.09 − 417.11 − EWW SpP 616.52 + 512.72 + 459.89 + 411.93 + EWW SuC 612.22 − 538.41 − 478.05 − 416.48 + EWW AuC 623.20 − 514.74 − 453.47 − 385.39 − EWW WiC 620.91 + 500.01 + 477.33 + 418.44 + EWW SpC 620.19 + 541.21 + 477.69 + 384.69 + WS WiP 633.58 + 548.20 + 470.59 − 393.62 − WS SpP 631.19 − 544.66 − 478.24 − 419.97 − WS SuC 633.68 − 538.62 − 472.98 − 413.44 − WS AuC 621.47 − 506.48 − 443.73 − 365.82 −0.251 (±0.038) 39.7% WS WiC 601.96 + 506.36 + 476.85 − 408.59 − WS SpC 630.54 − 512.11 − 435.49 − 417.43 − WD WiP 633.86 + 537.78 + 478.62 − 419.84 − WD SpP 608.04 − 538.40 − 475.80 − 420.38 + WD SuC 632.74 − 535.93 − 478.06 − 419.45 − WD AuC 607.96 − 531.57 − 466.50 − 417.62 − WD WiC 612.79 + 468.65 + 469.00 + 396.52 + WD SpC 599.17 − 479.57 − 418.75 −0.025 (±0.003) 31.7% SST WiP 630.87 − 539.98 + 471.94 + 408.49 + SST SpP 628.09 − 550.24 + 475.23 + 406.54 + SST SuC 628.86 − 547.08 − 478.78 − 420.21 − SST AuC 596.16 + 506.89 + 438.15 + 377.11 + SST WiC 628.92 + 534.12 + 462.08 + 379.65 + SST SpC 611.45 − 541.23 − 478.80 − 419.37 + TGR WiP 576.75 + TGR SpP 611.28 + 548.73 − 470.99 − 415.62 − TGR SuC 617.92 + 549.56 + 476.85 − 420.72 + TGR AuC 631.39 − 519.03 − 444.78 − 396.10 − TGR WiC 592.68 + 542.85 + 468.84 + 420.27 + TGR SpC 587.36 + 537.82 + 467.75 + 417.85 + TGR WiP d 548.24 0.394 (±0.043) 12.4% WD WiC d 476.81 0.028 (±0.003) 23.0%

First stage

The best-fitting first-stage covariate for adult survival was mean east-west sea tem- perature gradient in the previous winter (TGR WiP, AIC = 576.75, Table 6.3), with the positive regression coefficient (± standard error) of 0.257 (± 0.030) indicating that an increase in the mean east-west sea temperature gradient across Bass Strait in the winter of the previous year was associated with an increase in adult survival in the current year. However, as before, this covariate showed an increasing linear

101 temporal trend over the study period (p < 10−5, Figure 6.2). To remove the effects of this trend, the residuals of this covariate were used in the modelling, that is the covariate was de-trended. The resulting model with the de-trended covariate (TGR WiP d, Table 6.3) showed a marked improvement in AIC (AIC = 548.24).

The positive regression coefficient (± standard error) of 0.394 (± 0.043) indicated that increasing residuals of mean east-west sea temperature gradient in the pre- vious winter were associated with increased adult survival. In the context of this analysis, an increase in the residuals was an indication that the mean east-west sea temperature gradient was increasing relative to its mean value. Thus, an increase in adult survival was associated with a higher-than-average mean east-west sea tem- perature gradient in the winter of the previous year. The residuals of this covariate are shown in Figure 6.3.

Figure 6.11 displays the results of the best first-stage adult model for survival.

It shows the relationship between time-dependent adult survival φ2, indicated by either stars (for banded birds) or open circles (for transpondered birds), and the residuals of mean east-west sea temperature gradient in the previous winter. The dotted line is the model of predicted values for adult survival for banded birds and the solid line for transpondered birds when the residuals of mean east-west sea temperature gradient in the previous winter were used as the covariate. At the first stage, the proportion of deviance (time variation) explained was 12.4%.

Second stage

The best-fitting second-stage covariate was number of strong-wind days in the cur- rent winter (WD WiC, AIC = 468.65, Table 6.3), with the positive regression coef- ficient of 0.029 (± 0.005) indicating that an increase in the number of strong-wind days in the current winter was associated with an increase in adult survival. How- ever, as before, this covariate displayed a linear trend over the length of the study period (Figure 6.4). Hence the de-trended covariate was used in the modelling (WD WiC d, Table 6.3). This de-trended covariate produced an AIC which was lower than that of any other second-stage covariate (AIC = 476˙ .81, Table 6.3). The positive regression coefficient (± standard error) of 0.028 (± 0.003) indicated that, as the number of strong-wind days in the current winter increased relative to its mean, adult survival increased. The de-trended covariate residuals are shown in Figure 6.5. The correlation between the first-stage covariate and the de-trended covariate was low, an absolute value of 0.15, and at the second stage, the regres- sion coefficient (± standard error) of the first-stage covariate was 0.508 (± 0.045). The second-stage covariate increased the proportion of deviance (time variation) explained from 12.4% to 23.0%.

102 1

0.9

0.8

0.7

0.6

0.5

Adult survival 0.4

0.3

0.2

0.1

0 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 Residuals of mean east−west sea temperature gradient in previous winter

Figure 6.11: The relationship between time-dependent adult survival under model φ1(t+B) and residuals of mean east-west sea temperature gradient in the previous winter for banded (stars) and transpondered (open circles) birds. The lines show the predicted values for adult survival for banded (dotted line) and transpondered (solid line) birds for the fitted model of adult survival dependent on the residuals of mean east-west sea temperature gradient in the previous winter.

Third stage

The best-fitting third-stage covariate was number of strong-wind days in the current spring (WD SpC, AIC = 418.75, Table 6.3), with the negative regression coefficient (± standard error) of −0.025 (± 0.003) indicating that an increase in the number of strong-wind days in the current spring was associated with a decrease in adult survival. This covariate did not display any linear trend (p = 0.41) and had a low correlation with both the first- and second-stage covariates (absolute values of correlation coefficients 0.11 and 0.12, respectively). The regression coefficients (± standard error) of the first- and second-stage covariates in the third-stage model were 0.580 (± 0.046) and 0.026 (± 0.003), respectively. The third-stage covariate increased the proportion of deviance (time variation) explained by the model by a further 8.7%, from 23.0% to 31.7%.

Fourth stage

The best-fitting fourth-stage covariate was wind speed in the current autumn (WS AuC, AIC = 365.82, Table 6.3), with the negative regression coefficient (± stan-

103 dard error) of −0.251 (± 0.038) indicating that an increase in the wind speed in the current autumn was associated with a decrease in adult survival. This covariate did not display any linear trend (p = 0.50) and had a low correlation with the first-, second- and third-stage covariates (absolute values of correlation coefficients 0.07, 0.03 and 0.05, respectively). The fourth-stage covariate increased the proportion of deviance (time variation) explained by the model by a further 8%, from 31.7% to 39.7%.

6.3.4 Interaction between banding and the best covariate for adult survival

Figure 6.12 shows the interactive effect of the residuals of the mean east-west sea temperature gradient in the previous winter and banding on adult survival. The regression coefficients for both lines were positive and statistically significant for transpondered birds (p < 10−2) . The slope for banded birds was not statistically significant (p = 0.3).

0.86

0.84

transpondered birds

0.82

0.8 Adult survival 0.78 banded birds

0.76

0.74 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 Residuals of mean east−west sea temperature gradient in previous winter

Figure 6.12: The interactive relationship between adult survival for banded and transpondered birds and the residuals of the mean east-west sea temperature gra- dient in the previous winter.

6.4 Discussion

The best model for first-year survival included the following covariates: mean east- west wind speed in the previous spring; residuals of number of strong-wind days

104 in the current winter; mean sea-surface temperature in the previous winter; and mean east-west wind speed in the current autumn. The first three covariates had positive associations with first-year survival, the fourth a negative association. The proportion of total (time) deviance explained by this model was 26.5%. The best first-stage covariate was particularly significant for transpondered first-year birds.

The best model for adult survival included the following covariates: residuals of mean east-west sea temperature gradient in the previous winter; residuals of num- ber of strong-wind days in the current winter; number of strong-wind days in the current spring; and wind speed in the current autumn. The first two covariates had positive associations with adult survival, the second two negative associations. The proportion of total (time) deviance explained by this model was 39.7%. The best first-stage covariate was particularly significant for transpondered adult birds.

To interpret these results, I considered possible links between little penguin survival and the model covariates, based on the birds’ different patterns of movement.

Patterns of movement at sea of little penguins

Winter is a time when adult penguins are not constrained to land for breeding or moult. The mean duration of trips during this time is greater than at any other time of the year [Collins et al., 1999]. In July 1992, radio-tracked adult penguins from Phillip Island made foraging trips lasting on average 13.7 days (based on n = 16 observations with a standard error of 3.6 days, 44% single-day trips), and in July 1993, the same study reported foraging trips lasting on average 8.6 days (based on n = 38 observations with a standard error of 1.8 days, 61% single-day trips) [Collins et al., 1999]. In a subsequent study, McCutcheon et al. [2011] found that in winter, 72.3% of foraging trips made by satellite-tracked little penguins were single-day trips away from the colony (based on n = 47 foraging trips). The authors found that for birds making longer trips, the mean duration of trips was 22.7 days (based on n = 13 with a standard error of 4.3 days).

In the spring months, there is an increase in the number of birds coming ashore for the initial stages of breeding [Reilly and Cullen, 1981]. At this time, birds conduct territorial and courtship displays, burrow maintenance, and lay and incubate eggs [Reilly and Cullen, 1981, Stahel and Gales, 1987]. Recoveries of dead adult birds during this time were mainly in Port Phillip Bay and along the coastal region between Port Phillip Bay and Cape Otway [Dann et al., 1992] (Figure 6.13), with spring and autumn being periods of peak adult mortality [Dann et al., 1992]. Adult penguins are known to spend time before and during the breeding season in Port Phillip Bay, and remain within 20 km of the coastline most of the time, with a maximum daily range of around 30 km [Collins et al., 1999]. Satellite-tracking has

105 shown that in the period at the end of spring and beginning of summer, which coincides with incubation and the chick-guard stages of breeding, adult penguins make single-day trips into waters to the southwest of Phillip Island [Reilly and Cullen, 1981, Hoskins et al., 2008].

Little penguins on Phillip Island undergo catastrophic moult after breeding, with numbers of moulting birds reaching a peak at the end of summer/early autumn [Chambers and Dann, 2015]. The post-breeding/pre-moult interval ranges from less than four weeks to six weeks, depending on when birds ended their breeding [Reilly and Cullen, 1983]. During this time the birds build up reserves, putting on 30 − 40% of their body weight in preparation for the moult fast [Gales et al., 1988]. Little penguins are constrained to land during moult [Reilly and Cullen, 1983], unlike some seabirds which moult largely at sea, away from breeding grounds.

In contrast to adult birds, penguin chicks leave the colony immediately upon fledg- ing in the summer months and move in a westward direction, returning only to moult, around 12 – 14 months later. Their dispersal is indicated by recoveries of dead first-year birds banded on Phillip Island. Of the dead first-year birds recov- ered, around 37% were recovered away from the colony, predominantly concentrated along 400 km of coast running northeast and northwest from Cape Otway (see Fig- ure 6.1 for location) [Reilly and Cullen, 1982]. First-year birds from Phillip Island were recovered dead mainly in the first six months of life (January to June), pre- dominantly from late autumn to early winter (May to June) [Dann et al., 1992]. A similar pattern of movement, in which fledglings moved rapidly away from the colony while adult birds remained close to their natal colony, was observed in African penguins [Randall et al., 1987] and Gal´apagospenguins [Boersma, 1974], as well as in other seabirds such as the black-footed albatross Phoebastria nigripes [Gutowsky et al., 2014].

Figure 6.13 shows the numbers of recoveries of juveniles (aged 0−1 year) and adults (aged three years and over) banded as chicks on Phillip Island. Distinct differences in the recovery areas can be seen for the two age categories. Recoveries of first-year birds were widely dispersed, ranging from the coast between Port Phillip Bay and Cape Otway, and northwest from Cape Otway along the coast for about 800 km, with the majority found within about 300 km northwest of Cape Otway. Adult recoveries were concentrated around Phillip Island and in Port Phillip Bay [Dann et al., 1992, Reilly and Cullen, 1982]. Recoveries of birds marked with transponders are generally not reported by the public, as a specialised tag reader is required. Consequently I was unable to generate a similar figure using the data in my current analysis; dead-recovery data were generally unavailable for transpondered birds, unlike flipper-banded birds.

106 The clear spatial segregation at sea between first-year and adult birds may be due to active exclusion of first-year birds by adults from feeding grounds close to the colony, to different foraging abilities or diets of the two age classes or to different social requirements that limit time away from the colony for adult birds [Marchetti and Price, 1989, Dann, 1992, Bertellotti et al., 2002]. The vastly different patterns of movement at sea of first-year and adult birds led me to expect that the influences of the model covariates on survival would be via different mechanisms for the two age classes.

100 0 to 1 years

50

0

100 3 years and over

50

0 Ka 1 2 3 4 5 6 7 8 9 10 11PPBPh Fr 12 13 14 15 Ki Tas

Figure 6.13: The numbers of recoveries of juveniles (aged 0–1 years) and adults (aged three years and over) banded as chicks on Phillip Island. Each region repre- sents a 100 km section of the Victorian or South Australian coastline. The recoveries from Port Phillip Bay (PPB), Phillip Island (Ph), French Island (Fr), King Island (Ki), Kangaroo Island (Ka) and mainland Tasmania (Tas) are considered separately. Source: Sidhu et al. [2007] based on Dann et al. [1992].

107 Wind and sea temperature

In winter, sustained westerly and southwesterly winds that occur in Bass Strait generate eastward-moving currents which enter Bass Strait from the west [McInnes and Hubbert, 2003]. Historic drift-bottle experiments confirmed the eastward flow of waters in the northern region of Bass Strait [Olsen and Shepherd, 2006]. Sandery and K¨ampf[2005] suggested that the winter-spring flushing of Bass Strait results from eastward advection of the South Australian Current (SAC) and the Subantarc- tic Water (SAW), replenishing the Strait waters in a period of approximately 30 days in conditions of strong westerly winds (Figure 6.1). The easterly advection into Bass Strait of the colder waters of the SAC and the SAW potentially support higher productivity and affect prey abundance [Gibbs, 1992]. Furthermore, intermittent strong winds over stratified waters can cause a temporary deepening of the thermo- cline, causing nutrient-rich waters to be mixed upwards. This can result in a burst of phytoplankton production, with a subsequent increase in fish stocks, an occur- rence that was noted in eastern Tasmania by Mann [1993]. In southern Australia, the prevalence of winds during the austral summer and autumn facilitated the in- trusion of cold water onto the continental shelf in southern Australia, resulting in greater levels of eggs and larvae of sardine Sardinops sagax and anchovy Engraulis australis [Ward et al., 2006], both of which are major prey species of little penguins [Montague and Cullen, 1987, Cullen et al., 1992]. Strong westerly winds blowing in the winter accelerated the movement of cooler, nutrient-rich waters into penguin feeding areas, resulting in earlier laying, heavier birds and higher breeding success four months later [Mickelson et al., 1992]. In contrast, increased westerly winds led to an increase in foraging effort by little penguins on Gabo Island in southeastern Australia, suggesting that the wind prevented the flow of nutrient-rich water from the south, resulting in reduced prey availability [Berlincourt and Arnould, 2015].

The confluence of water masses which influences Bass Strait also affects the sea- surface temperature of the region [Sandery, 2007] (Figure 6.1). An increase in an east-west sea temperature gradient may mean cooler waters entering western Bass Strait or warmer waters in eastern Bass Strait. The southward flow of currents along the east coast of Australia is projected to strengthen and to continue to warm by 1−2◦C by the 2030s and by up to 3◦C by the 2070s; eastern Bass Strait is therefore predicted to warm more than western Bass Strait [Poloczanska et al., 2007]. As a result, movement of fish, and the timing of peak productivity of spawning and juvenile prey species of little penguins, are expected to shift forward in time in the southern regions of Australia [Cullen et al., 2009]. Additionally, ocean warming in southeastern Australia is causing poleward shifts in a range of species [Poloczanska et al., 2007], including species on which penguins and their prey depend.

108 Linking survival and model covariates

The covariate which accounted for the highest proportion of the total (time) de- viance explained for first-year survival was mean east-west wind in the previous spring, with a positive association with first-year survival. (Note that an increase in this covariate implies increasing wind from the west, Table 6.1.) At this time, chicks were not yet alive. Thus, this suggested a mechanism which resulted in chicks beginning their first year of life with a higher probability of surviving that year. This mechanism could be via their parents, or by other unknown variables which affected later conditions such as food availability after the chicks fledged. Spring marks the period when more birds come ashore for the initial stages of breeding; during this time, adult birds generally forage within 30 km of the colony. An in- crease in prey availability in waters surrounding the colony at this time, facilitated by westerly wind-assisted movement of nutrient-rich water into this region, would lead to heavier adults which could then confer this advantage on their chicks, re- sulting in increased first-year survival.

Evidence of the effects of winter carry-over body mass on subsequent breeding per- formance of little penguins was found by Salton et al. [2015]; birds with higher body mass began laying earlier, and heavier males had higher breeding success. Post-fledging survival of little penguins was influenced by weight at fledging, with heavier chicks more likely to survive than lighter ones [Reilly and Cullen, 1981, 1982, Knight and Rogers, 2004]. If increased body condition of adult little penguins in the pre-breeding and incubation periods led to earlier breeding and higher fledging weights [Nisbet and Dann, 2009, Robinson et al., 2005, Cullen et al., 1992], condi- tions which influence adult body condition will have an indirect positive effect on chick survival. Other studies such as Chiaradia and Nisbet [2006] have linked longer parental foraging-trip duration (an indication of depressed food availability) with lower growth parameters of little penguin chicks. Southern Rockhopper penguins bred earlier and laid heavier eggs under cooler conditions (reflecting higher primary productivity) brought about by lower sea-surface temperatures and stronger west- erly winds [Dehnhard et al., 2015a,b], while little penguins were heavier and bred earlier when strong westerly winds blew four months earlier [Mickelson et al., 1992]. Hence, it is conceivable that favourable conditions in the period immediately before and during the early part of the breeding season have a positive impact on a pen- guin’s first year of life. My results here have provided evidence of a link between an increase in survival in the first twelve months of a little penguin’s life and increased parental body condition just before and at the start of the breeding season the year before, linked to increasing winds from the west at those times.

109 The second-best covariate for first-year survival was the residuals of number of strong wind days in the current winter (the de-trended covariate), with a positive association with first-year survival. This covariate was also positively associated with adult survival at the second stage. An increase in this covariate was an in- dication that the number of strong-wind days in the current winter was increasing relative to the mean. Clearly this covariate affected first-year and adult birds via different mechanisms; winter is before the breeding season and so the chicks were not yet born. Furthermore, first-year birds had not yet returned to the colony to moult. Adult birds remained closer to the colony as winter progressed, in prepa- ration for the breeding season. An increase in local seasonal winds, represented by this covariate, probably provided the mechanism by which primary productivity was enhanced in foraging grounds. Thus adult birds were able to build up body condition before the breeding season, with the subsequent positive impact on both first-year and adult survival.

Two sea-temperature covariates were found to positively affect penguin survival: SST in the previous winter at the third stage for first-year birds; and the residuals of east-west sea temperature gradient, also in the previous winter, at the first stage for adult birds. The positive lagged effect of this covariate on first-year survival again suggests that its effect was via parental investment in young birds during breeding, or that it led to better conditions for first-year birds several months later. Other studies have related the effects of SST to altered little penguin breeding variables. For example, warm water that occurred up to six months prior to the onset of breeding was associated with improved breeding success of little penguins on Phillip Island [Cullen et al., 2009]. High SST in the summer immediately pre- ceding breeding was associated with less-successful breeding, and cooler SST in the winter of the year before breeding was positively correlated with an earlier start to breeding on Penguin Island in Western Australia [Cannell et al., 2012]. Karnauskas et al. [2015] linked an increase in the population of Gal´apagospenguins Speniscus mendiculus to cooler local SST, brought about by an increase in local upwelling due to decreasing trade winds.

Furthermore, adult survival was positively associated with the residuals of east-west sea-temperature gradient in the previous winter. An increase in the residuals was an indication that the east-west sea-temperature gradient was increasing relative to its mean value over the length of the study. This implied that either the waters in western Bass Strait were cooler than usual or that the waters in eastern Bass Strait were warmer than usual, or both. The lagged association of this covariate with annual survival suggests that conditions influenced adult birds in the pre-breeding season such that they had a higher probability of survival in the year immediately

110 following the breeding season. For adult little penguins, the breeding season (spring and summer) is followed by the moult period (autumn). Birds with better body condition breed earlier and tend to moult earlier [Reilly and Cullen, 1983]; perhaps successful breeders are also more efficient foragers and are able to build up enough body reserves to survive the days of fasting during moult [Reilly and Cullen, 1983]. Von Br¨omssenand Jansson [1980] found that earlier breeding allowed more time to improve body condition before moult and was related to better adult survival.

First-year survival was negatively associated with mean east-west wind speed in the current autumn at the fourth stage. As most mortality of first-year little penguins occurs in the period from late autumn to early winter [Dann et al., 1992], this may be a consequence of the high energy expenditure during foraging by inexperienced first-year birds. The best third- and fourth-stage covariates for adult survival were ones which included the effect of wind speed and the number of strong-wind days, both with negative associations with adult survival, and both during periods of peak adult mortality in spring and autumn [Dann, 1992]. Strong winds have also been linked with other severe weather conditions such as rough seas, which could affect prey dispersion and penguin foraging patterns [Berlincourt and Arnould, 2015, Saraux et al., 2016]. Little penguins are known to avoid rough sea conditions and remain ashore until conditions ease [Weavers, 1992]. Thus, it is conceivable that wind conditions at these times have a negative association with survival of both first-year and adult little penguins.

The effects of wind on penguin survival are complex. The severity of strong-wind events is projected to increase in the future, with regional winds predicted to weaken and move poleward, while the severity of storms is projected to increase [Poloczan- ska et al., 2007]. Increasing wind can have both positive and negative effects on penguin survival via several mechanisms. For example, reduced foraging due to rough sea conditions, and wind-induced water turbulence resulting in the disper- sion of prey can negatively affect survival. However, wind can also facilitate the movement of nutrient-rich waters, which support secondary production, to penguin feeding grounds, thus having a positive impact on survival via prey recruitment.

This chapter investigated possible relationships between first-year and adult little penguin survival, and a range of wind and sea-temperature covariates. By using both wind and sea-temperature covariates in the same model for little penguin survival, I was better able to predict survival than if either wind or sea temperature was used alone. The effects of these covariates on penguin survival were complex and not always immediate, with lagged covariates found to affect both first-year and adult survival. These findings allow us to gain insight into some of the processes occurring at sea that may affect little penguin survival.

111

Chapter 7

Marine productivity and survival of little penguins — implications of a seasonal coastal upwelling

The aim of this chapter is to investigate how marine productivity in southeastern Australia affects first-year and adult survival of little penguins. Chlorophyll a concentration is used here as a proxy for marine productivity. The results from this chapter form the basis of a manuscript entitled ‘Marine productivity and survival of first-year and adult little penguins Eudyptula minor in southeastern Australia — implications of a seasonal coastal upwelling’ [Ganendran et al., in prep.].

7.1 Introduction

Food supply is a significant factor in the regulation of population sizes of seabirds and several mechanisms have been proposed [Lack, 1954, 1968, Cairns, 1989, 1992]. Ashmole’s ‘halo effect’ model of population regulation proposed that large aggrega- tions of animals may deplete food resources near the colony [Ashmole, 1963]. In a refinement of Ashmole’s [1963] hypothesis, Furness and Birkhead [1984] postulated that seabird colony sizes were inversely proportional to the number of conspecifics from other colonies competing for food near the colony in common feeding grounds (the ‘hungry-horde’ model). An alternative model, the ‘hinterland’ model proposed by Cairns [1989], suggested that seabirds from neighbouring colonies occupy non- overlapping foraging areas, and that the size of the colony is a function of the size of the foraging area.

In marine environments, food availability is dependent on oceanographic conditions, interspecific and intraspecific competition and extraction pressure by fisheries [Fur- ness, 2003, Mann and Lazier, 2006, Cury et al., 2011]. Seabird foraging behaviour is strongly influenced by the temporal and spatial availability of prey [Montevecchi and Myers, 1995, Monaghan, 1996, Bost et al., 2009, Sommerfeld et al., 2015], which

113 are often concentrated into patches associated with mesoscale oceanographic fea- tures such as upwellings, fronts or eddies [Boersma et al., 2009, Harding et al., 2013]. Inshore foraging seabirds adjust their foraging strategies in response to variation in food supply particularly during periods of prey density fluctuations, for example, by foraging preferentially in eddies where prey biomass was likely to be higher [Cott´e et al., 2007], adjusting foraging trip distances to correspond with areas of high pro- ductivity [Boersma et al., 2009], and using a variety of dive patterns in response to prey availability [Lescro¨eland Bost, 2005]. Low levels of food would be expected to affect future fitness [Davis et al., 2005, Kitaysky et al., 2010]. For example, fe- cundity [Monaghan et al., 1989, Gill et al., 2002, Mills et al., 2008] and deferred breeding have been shown to be linked to food availability [Boersma, 1977, 1978, Crawford and Dyer, 1995, Mitchell et al., 2004]. In addition, Af´anet al. [2015] provided evidence that little penguins at Phillip Island were able to adjust their reproductive timing to synchronise with the annual peak of marine productivity.

Little penguins are inshore foragers, generally remaining within 20 km of the coast, with a maximum average daily range of around 30 km in the breeding season, and less frequent longer trips further west and northwest in the winter [Collins et al., 1999, McCutcheon et al., 2011]. During the breeding season, in addition to daily maintenance energy requirement, adult birds are faced with the constraint of having to return to the colony to feed their offspring [Costa, 1991], and as food supply depletes across the season, adaptive foraging strategies such as changes in foraging trip length, are required to continue provisioning offspring and for self-maintenance [Lewis et al., 2006, Hoskins et al., 2008, Saraux et al., 2011, Berlincourt and Arnould, 2015].

In contrast, chicks leave the colony immediately upon fledging in the summer months and move in a westward direction (refer to Section 6.4 for a detailed dis- cussion on the different patterns of movement of little penguins at sea). The clear spatial segregation at sea between first-year and adult birds may be due to active exclusion of first-year birds by adults from feeding grounds close to the colony, different foraging abilities or diets of the two age classes, or different social require- ments that limit time away from the colony for adult birds [Marchetti and Price, 1989, Dann, 1992, Bertellotti et al., 2002]. A similar pattern of movement in which fledglings moved rapidly away from the colony while adult birds remained close to their natal colony was observed in African penguins Spheniscus demersus [Randall et al., 1987] and Gal´apagospenguins Spheniscus mendiculus [Boersma, 1974], as well as in other seabirds such as the black-footed albatross Phoebastria nigripes [Gutowsky et al., 2014].

114 In this study, I investigated the effect of marine productivity on survival of first-year and adult little penguins by using chlorophyll a as a proxy for productivity. There were two regions of importance for adult and first-year penguins respectively: the waters immediately surrounding Phillip Island, and the more distant waters several hundred kilometres west of Phillip Island where a seasonal upwelling occurs. The covariates used in survival models were created to reflect the likely foraging areas of little penguins.

7.2 The coastal upwelling process and the Bonney Upwelling

7.2.1 Chlorophyll a

Chlorophyll a is generally accepted as an indicator of phytoplankton biomass, and changes in biomass generally reflect the changes in primary productivity [Thurman and Trujillo, 2004]. Globally, chlorophyll a has decreased at a rate of around 1% per year through the 20th century [Boyce et al., 2010]. Recent ocean warming has driven change in productivity, to the extent that global circulation models predict further reductions in the 21st century [Thomas et al., 2012]. However, in some coastal regions (depth < 200 m) such as the Patagonian and California/Mexican Shelves, chlorophyll a levels have increased significantly, possibly due to a decrease in sea temperature, increases in land-based nutrient inputs or an increase in upwelling caused by a warming Earth [Gregg et al., 2005]. In Bass Strait, southeastern Aus- tralia (Figure 7.1), the region of interest in this study, an overall increasing trend in chlorophyll a was reported after 2005, with an increase during autumn and winter and a decrease in spring and summer [Kelly et al., 2015].

Wind-induced upwelling is a major mechanism for bringing nutrients to the ocean surface. In coastal upwelling systems, the driving force is wind stress from equator- ward winds [Mann and Lazier, 2006]. The wind-induced movement of surface wa- ters drags the layer beneath it, setting in motion other deeper layers. Coastal upwelling occurs when this surface water is replaced by cooler waters. The newly surfaced water is often rich in nutrients, so that prolonged winds result in increased biological productivity [Thurman and Trujillo, 2004, Garrison, 2005]. Upwelling systems at the eastern boundaries of the oceans of the world are highly productive nutrient-rich environments, accounting for more than 20% of the world’s marine catch [Rykaczewski and Checkley, 2008] while occupying less than 1% of the ocean surface area [Botsford et al., 2006].

115 7.2.2 The Bonney Upwelling

The Bonney Coast is an area spanning around 800 km between Cape Jaffa, South Australia (36◦580S, 139◦420E) and Portland, Victoria (38◦210S, 141◦360E) [Schahinger, 1987, Lewis, 1981]. The continental shelf from Cape Jaffa to Portland is narrow, around 25 − 55 km wide, and broadens out to a width of around 50 − 80 km east of Portland. The Bonney Upwelling is a large seasonal wind-driven coastal upwelling that starts around November and continues to April (austral late spring to autumn) [K¨ampfet al., 2004] along the Bonney Coast. The cold upwelled waters are associ- ated with increased nutrient concentrations that support elevated levels of primary production, providing food for large numbers of seabirds, including little penguins [Reilly and Cullen, 1982, Dann et al., 1992] and, among others, Australian Arcto- cephalus pusillus and long-nosed A. forsteri fur seals [Baylis et al., 2008], southern bluefin tuna Thunnus maccoyii [Ward et al., 2006] and the endangered blue whale Balaenoptera musculus [Gill et al., 2011]. The cold-water plume associated with the upwelling event is clearly visible on the sea-surface temperature satellite map taken in March 1995 (Figure 7.1) [Butler et al., 2002].

Upwelling along the Bonney Coast has been shown to be associated with localised increases in surface chlorophyll a and downstream enhancement of zooplankton biomass [Gill et al., 2002, K¨ampfet al., 2004, Nieblas et al., 2009]. High densities of eggs and larvae of sardine Sardinops sagax and anchovy Engraulis australis, both prey species of little penguins [Cullen et al., 1992, Chiaradia et al., 2012], occur in areas with high zooplankton biomass [Ward et al., 2006]. Ward et al. [2006] also suggested that southern bluefin tuna migrate to the central and eastern areas of the Great Australian Bight, where high densities of sardines are available during the upwelling period. Lactating long-nosed fur seals shift their foraging location in response to seasonal availability of prey, attributed to productivity associated with the Bonney Upwelling [Baylis et al., 2008].

7.3 Data and methods

7.3.1 Penguin data

The time series of penguin data used in this analysis was an abbreviated subset of the main dataset described in Section 2.3.5 on page 27 to correspond to the number of available years of chlorophyll a data. The reduced dataset (1999 − 2012) comprised 5,159 banded birds (2,935 banded as chicks close to fledging and 2,224 banded as adults of unknown age) and 10,162 transpondered birds (6,977 transpondered as chicks close to fledging and 3,185 transpondered as adults of unknown age). Due to the sharp decrease in the number of chicks banded after

116 Figure 7.1: Satellite image from March 1995 showing sea-surface temperature, in particular the strong cold-water upwelling along the Bonney Coast indicated by the area of dark blue [Butler et al., 2002].

2002, when marking shifted to transponders, the results for first-year survival are reported only for transpondered birds.

7.3.2 Chlorophyll a data

Chlorophyll a data used in this analysis were described in Section 2.6.4.

7.3.3 Study regions

First-year birds, all of pre-breeding age [Dann and Cullen, 1990], move westward away from the Phillip Island colony to feeding grounds several hundred kilometres along the coast, as far as Cape Jaffa (Figure 7.2), and return to the colony for their first moult around a year later [Reilly and Cullen, 1982, Dann et al., 1992]. Most adult birds go westward during foraging trips but remain close to the colony in the pre-breeding and breeding seasons, making frequent trips back for burrow renovation, pair maintenance and for incubation and feeding of chicks [Sidhu et al., 2007]. They move further away, also in a westward direction, during non-breeding periods but generally stay within 15 km of the shore when foraging [Dann et al., 1992, McCutcheon et al., 2011]. This difference in patterns of movement for birds of different ages is highlighted in Figure 6.13, which shows the distribution of dead

117 recoveries for various regions along the Victorian and South Australian coastline [Sidhu et al., 2007].

The regions of interest in this study are labelled as Boxes 1 and 2 (Figure 7.2). Box 1 encompasses the area around Phillip Island and the coast west to Cape Otway, where adult birds are known to forage in the non-breeding and breeding periods. Box 2 encompasses the area between Cape Otway (38◦510S, 143◦310E) and Cape Jaffa (36◦560S, 139◦410E), to which first-year birds are known to travel. The seaward edge of Box 2 roughly follows the continental shelf at the 200 m isobath. Adult little penguins usually stay within 15 km of the shore when foraging. Box 1 represents an area of approximately 200×160 km, Box 2 approximately 400×100 km.

Figure 7.2: Regions of chlorophyll a considered in this study, labelled as Boxes 1 and 2.

7.3.4 Model covariates

In this study, the small number of years of chlorophyll a data was a limitation that could potentially have led to spurious results [Grosbois et al., 2008]. To reduce the likelihood of this, I restricted modelling to reflect immediate temporal and spatial effects of chlorophyll a on survival [Catchpole et al., 2000]. Hence models were limited to two stages and to covariates from the current year only.

Based on the likely foraging areas of little penguins from Phillip Island and recovery patterns of dead birds, I considered Boxes 1 and 2 for first-year birds. I have tried

118 to simplify my study to focus on the direct effects of marine productivity in the predominant area where adult birds forage. As adult birds spend most of their lives in the region defined by Box 1, I considered only Box 1 when modelling adult survival.

The covariates were mean chlorophyll a in each of the boxes shown in Figure 7.2 over a 12-month period, encompassing four seasons in one calendar year. The covariates for survival in calendar year j had the suffixes as previously described for each season, that is from the Current Summer (1 December in calendar year j −1 to 28 or 29 February in calendar year j ), denoted by ‘SuC’, to Spring in the Current year (1 September – 30 November in calendar year j ), denoted by ‘SpC’ (see also Table 6.1 on page 92 for definition of suffixes). The prefixes ‘Box1’ and ‘Box2’ correspond to the regions shown in Figure 7.2. For example, ‘Box1 SpC’ was the mean chlorophyll a in Box 1 in the current spring.

7.3.5 Methods

Methods, including model fitting and notation, and the formation of the likelihood, are described in detail in Chapter 3. Specifically, the interactive effects of the co- variates and age on survival are described in Section 3.6. As non-linear relationships have previously been found between survival and food supply [Cairns, 1988, Piatt et al., 2007], I also fitted quadratic models for first-year and adult survival in the th 2 j calendar year of the general form logit(φj) = a + bXj + cXj where Xj is the covariate value in the jth calendar year. As in earlier chapters, all survival models for adult survival also included a banding effect.

7.4 Results

Tables 7.1 and 7.2 contain values of the Akaike’s information criterion (AIC) for models using seasonal chlorophyll a covariates for first-year and adult survival prob- abilities respectively. The AIC of the null model for first-year survival was 226.98, while the AIC of the null model for the adult survival was 57.48. The null model for first-year survival has first-year survival depending only on whether or not a bird was banded (covariate B) and adult survival fully time-varying and depending on whether or not a bird was banded; similarly for the null model for adult survival. The AIC of the full model, in which both first-year and adult survival probabilities were fully time-varying (and dependent on B), was 17.06.

119 7.4.1 First-year survival

The best-fitting first-stage model for first-year survival was a quadratic function of the mean chlorophyll a in the region of the Bonney Upwelling in the current sum- mer (Box2 SuC quad, Table 7.1). The negative quadratic coefficient indicated that an increase in mean chlorophyll a in this region was associated with an increase in first-year survival to a threshold value of the covariate, after which survival de- creased. This model accounted for 33.2% of the proportion of deviance (time vari- ation) explained. Compare this with the best first-stage linear model (Box1 SuC, AIC = 188.26, Table 7.1) which accounted for 17.4% of the proportion of deviance (time variation) explained (result not shown). The first-stage covariate displayed no linear trend over time (p = 0.65).

Table 7.1: Akaike’s information criterion (AIC) values for models with the covari- ates listed for first-year survival probability. The suffix ‘ quad’ indicates a quadratic function of the covariate listed. All models used an (age + time) dependence for re- capture probabilities, and adult survival was fully time-varying (and depended on whether or not birds were banded, B). The second-stage model used the best model from the first stage with an additional covariate. The proportion of time deviance explained by the best model at each stage is also given, in bold.

Stage 1 Stage 2 Covariate AIC AIC Box1 SuC 188.26 97.38, 58.0% Box1 AuC 194.82 109.85 Box1 WiC 206.2 123.06 Box1 SpC 214.48 148.02 Box2 SuC 207.15 Box2 AuC 194.52 140.50 Box2 WiC 213.01 150.58 Box2 SpC 218.29 136.54 Box1 SuC quad 169.13 108.63 Box1 AuC quad 192.82 118.87 Box1 WiC quad 207.91 129.02 Box1 SpC quad 214.62 150.16 Box2 SuC quad 153.29, 33.2% Box2 AuC quad 194.91 140.83 Box2 WiC quad 195.65 148.91 Box2 SpC quad 213.91 136.92

120 The best-fitting second-stage covariate for first-year survival was the mean chloro- phyll a in the waters around Phillip Island in the current summer (Box1 SuC, Table 7.1), with the negative regression coefficient (± standard error) of −0.774 (± 0.156) (result not shown) indicating that an increase in mean chlorophyll a in the summer in this region was associated with a decrease in first-year survival. This covariate substantially increased the proportion of deviance (time variation) explained by the model to 58.0%.

At the second stage, the quadratic regression coefficient of the best first-stage co- variate remained virtually unchanged at −0.028 (± 0.008) (result not shown). The second-stage covariate displayed no linear trend over time (p = 0.23). The correla- tion between the first- and second-stage covariates was low, with an absolute value of the correlation coefficient of 0.18.

0.35

0.3

0.25

0.2

0.15 First−year survival

0.1

0.05

0 10 12 14 16 18 20 22 24 Mean chl−a concentration in Box 2 in current summer (mg/m3)

Figure 7.3: The relationship between first-year survival under the fully time-varying model and mean chlorophyll a in the region of the Bonney Upwelling (Box 2) in the current summer (open circles). The curve shows predicted values for the fitted quadratic model of first-year survival dependent on the mean chlorophyll a in Box 2 in the current summer.

7.4.2 Adult survival

The best-fitting first-stage covariate for adult survival was the mean chlorophyll a in the waters around Phillip Island in the current summer (Box1 SuC, Table 7.2, Figure 7.4), with the positive regression coefficient (± standard error) of 0.219 (±

121 0.063) (result not shown) indicating that an increase in the mean chlorophyll a in the current summer in this region was associated with an increase in adult survival. This covariate accounted for 15.7% of the proportion of deviance (time variation) explained. It displayed no linear trend (p = 0.23). I also needed to consider a competing model containing the quadratic form of this covariate as it had an AIC value of 50.92, a difference of <1 from the model containing the linear covariate (AIC = 49.39, Table 7.2). However, the positive sign of the quadratic coefficient indicated a curve that was concave upwards which implied that as marine produc- tivity increased, survival decreased to a minimum point and then increased again. From a biological perspective, this is not meaningful as adult survival would be expected to increase rather than decrease as marine productivity increased [Cairns, 1988]. Therefore I chose the linear model to be the best-fitting first-stage model.

Table 7.2: Akaike’s information criterion (AIC) values for models with the covari- ates listed for adult survival probability. The suffix ‘ quad’ indicates a quadratic function of the covariate listed. Adult survival also depended on whether or not birds were banded, B. All models used an (age + time) dependence for recapture probabilities, and first-year survival was fully time-varying. The second-stage model used the best model from the first stage with an additional covariate. The propor- tion of time deviance explained by the best model at each stage is also given, in bold.

Stage 1 Stage 2 Covariate AIC AIC Box1 SuC 49.39, 15.7% Box1 AuC 58.46 39.98 Box1 WiC 59.37 39.92 Box1 SpC 58.20 48.35 Box1 SuC quad 50.92 50.91 Box1 AuC quad 57.54 41.34 Box1 WiC quad 61.17 32.38, 45.2% Box1 SpC quad 59.11 48.84

The best-fitting second-stage covariate for adult survival was a quadratic function of the mean chlorophyll a concentration in the waters around Phillip Island in the current winter. The negative sign of the quadratic coefficient indicated a curve that was concave downwards, i.e. increasing chlorophyll a was associated with increasing adult survival to a threshold value of the covariate after which survival decreased. The second-stage covariate substantially increased the proportion of deviance (time variation) explained by the model from 15.7% to 45.2%.

122 0.9

0.85

0.8

0.75

0.7

0.65 Adult survival

0.6

0.55

0.5

0.45 0.5 1 1.5 2 2.5 3 Mean chlorophyll a concentration in Box 1 in current summer (mg/m3)

Figure 7.4: The relationship between adult survival under the fully time-varying model and mean chlorophyll a in Box 1 in the current summer for banded (stars) and transpondered (open circles) birds. The lines show the predicted values for adult survival for banded (dashed line) and transpondered (solid line) birds for the fitted model of adult survival dependent on the mean chlorophyll a in Box 1 in the current summer.

The covariates used for the adult survival models were those for the waters im- mediately surrounding Phillip Island, with high correlations between them (see Tables C.8 and C.9 in Appendix C). The Principal Component Analysis (PCA) for these covariates (Section D.2.4, p. 182) showed that the first three covariates were moderately strongly positively correlated, and the first PC (PC1) accounted for approximately 48% of the variance in the data. The linear combination of this PC1 (p. 182) indicates that it represents high levels of chlorophyll a in the waters around Phillip Island in the summer, autumn and winter. In a practical sense, an increase in this PC would have a positive effect on adult survival; however it is difficult to determine which of the covariates had the greatest effect on survival. Furthermore, using PC1 as a covariate in the adult survival model yielded a higher AIC (58.66) than the best-fitting first-stage covariate.

7.4.3 Interaction between banding and the best covariate for adult survival

Survival models for the interactive effects of banding and the mean chlorophyll a in the waters around Phillip Island (Box 1) in the current summer on adult survival are shown in Figure 7.5. The slopes of both lines were statistically significant (p < 10−2 for banded birds and p = 0.03 for transpondered birds).

123 0.9

0.88

0.86 transpondered birds 0.84

0.82

0.8 Adult survival 0.78

0.76 banded birds 0.74

0.72

0.7 0.5 1 1.5 2 2.5 3 Mean chlorophyll a concentration in Box 1 in summer (mg/m3)

Figure 7.5: The interactive relationship between the effects of banding and the mean chlorophyll a in Box 1 in the current summer on adult survival.

7.5 Discussion

The best model for first-year birds showed a quadratic association between mean chlorophyll a values along the Bonney Coast in the current summer, indicating a positive association with first-year survival to a threshold level of the covariate be- yond which survival decreased. In addition to the amount of variation accounted for by this model (32.2%), some additional confidence can be derived from the gen- eral overlap both temporally and spatially of the first-year birds along the Bonney Coast at this time (summer). The first-year birds usually fledge from November to February [Reilly and Cullen, 1981] and their peak mortality period is highest in the first six months of life, with the majority of birds succumbing to starvation and/or parasitic disease [Dann et al., 1992, Harrigan, 1992]; hence this covariate would be expected to have greatest impact in this period. The quadratic model suggested a possible optimal chlorophyll a concentration for survival of first-year little penguins. Quadratic relationships between seabird demography and environmental variables are discussed below.

The best model for adult survival included the following covariates: the mean chloro- phyll a in the current summer in the waters in northwestern Bass Strait near Phillip Island and a quadratic function of the mean chlorophyll a in the same area in the current winter. Both of these covariates had positive associations with survival, i.e. the higher the chlorophyll a levels in summer, the higher the adult survival;

124 in winter, the higher the chlorophyll a levels, the higher the adult survival until a threshold level of chlorophyll a was reached. The proportion of deviance (time variation) accounted for by the model was 45.2%. The positive association for the best first-stage covariate was significant for both banded and transpondered adult birds.

The spatial segregation of first-year and adult birds is highlighted by the above- mentioned results. The breeding season on Phillip Island peaks with chick-rearing in summer [Reilly and Cullen, 1981] and is a time of high productivity, with a concentration of adult penguins foraging in waters close to the colony. In the breeding season, the colony may consist of up to 32,000 breeding adults [Sutherland and Dann, 2014], with their broods of 1 − 2 chicks. Adults make single-day trips during this time while provisioning chicks, and it is very likely that food depletion occurs within close proximity to the colony where most of the birds feed [Dann and Norman, 2006]. This is also a time when newly fledged chicks first go to sea, dispersing widely in a predominantly westward direction (towards the local major source of productivity along the Bonney Coast) away from an area near the colony potentially depleted of food and where they would likely face competition from adult birds for the remaining resources. This westward dispersion appears to be an adaptive strategy for inexperienced young birds to avoid high intraspecific competition with breeding adults; evidence of this is provided by the best second- stage covariate with a negative association between first-year survival and marine productivity in the waters around Phillip Island. With no breeding responsibilities, newly fledged chicks are able to exploit the distant feeding grounds rather than compete with adult birds for prey in near-colony waters. Waters to the west of Phillip Island support other smaller colonies of little penguins; for example at Port Campbell National Park (38◦380S, 143◦40E, approximately 60 km west of Phillip Island) the mean mass of birds is higher than that of breeding birds on Phillip Island [Dann et al., 1992], suggesting that this region may be a relatively rich feeding area with few breeding adults competing for food [Dann and Norman, 2006]. Hence the positive association between first-year survival and marine productivity in this region of the Bonney Upwelling in the summer and the negative association with marine productivity in waters around Phillip Island during the same season is not surprising. The second-stage model accounted for 58.0% of the proportion of deviance (time variation).

The circulation of Bass Strait is extremely complex and highly dynamic, with sea- sonal variation in flushing time, residence time and water-age distributions [Sandery, 2007]. In winter, tracers released at the northwestern corner of Bass Strait revealed

125 an eastward advection adhering to the Victorian coastline due to the South Aus- tralian Current (SAC, Figure 6.1) [Sandery and K¨ampf,2005]. Additionally, the cold and nutrient-rich Subantarctic Water reaches the western and eastern ends of Bass Strait in winter [Gibbs, 1992]. The eastward movement of water which results in winter flushing is a significant process for nutrient replenishment of the region. In summer, a weak westward flow is established, and eastern Bass Strait is influenced by warmer water carried by the East Australian Current (EAC, Figure 6.1) [Gibbs, 1992, Olsen and Shepherd, 2006]. Additionally, K¨ampf[2015] found evidence of a significant seasonal upwelling on the west Tasmanian shelf which appears in the late austral summer. Although it is not entirely clear how these nutrient-enriched peripheral waters mix into the waters around Phillip Island, it does appear that zooplankton biomass is higher in shallow water [Gibbs, 1992] such as in northern Bass Strait and around Phillip Island.

It is not surprising that an increase in marine productivity in the waters around the Phillip Island penguin colony during the latter part of the breeding season (summer) had a positive impact on adult penguin survival. The area around Phillip Island in a 30−40 km radius supports the food requirements of up to 45, 000 penguins (adults and chicks) during chick-rearing periods and it is likely that prey depletion occurs. For parents trying to balance investment in their offspring against their own chances to reproduce in the future [Stearns, 1976], an increase in marine productivity in these waters during this time would provide a buffer for penguins coping with the energetic demands of chick provisioning, a period of highest energy demand [Gales and Green, 1990], while trying to maintain their own fitness. Other seabird studies have shown the relationship between food availability and the energetic balancing of chick provisioning and parental body condition. For example, yellow- nosed albatrosses Diomedea chlororhynchos at Pointe d’Entrecasteaux, Amsterdam Island, are able to adjust their provisioning effort to deliver larger meals to chicks during favourable food conditions, but increase the duration of their foraging trips, probably to recover some of their own body condition, during times of low food availability [Weimerskirch et al., 2001]. Similarly, Antarctic petrels Thalassoica antarctica at Dronning Maud Land, Antarctica in good body condition were able to deliver more food to chicks, whereas those in poor body condition did not respond to their chick’s needs, probably to maximize their own chances of surviving [Tveraa et al., 1998].

Winter is a time of reduced prey availability for penguins at Phillip Island [Hob- day, 1992] when adult little penguins make longer trips into Port Phillip Bay [Mc- Cutcheon et al., 2011], while still remaining relatively close to the colony to attend to pre-breeding activities such as burrow maintenance and courtship. It may be that

126 individuals with greater winter body mass are better able to manage unpredictabil- ity in food supply, increasing their chances of survival that year. A relationship between body mass and survival has been shown in other seabird species; for exam- ple, Harding et al. [2011] found a positive relationship between body mass at the end of the breeding season and post-breeding survival in the little auks Alle alle, and Haramis et al. [1986] related early-winter body mass to the probability of winter and annual survival for the water fowl Canvasback Aythya valisineria. Salton et al. [2015] reported that, contrary to expectations, the winter body mass of male and female little penguins was equal to or higher than in other inter-breeding periods, leading to earlier laying and more successful breeding. A similar carry-over effect of winter body condition into the subsequent breeding season has also been found for other seabird species [Harding et al., 2011, Szostek and Becker, 2015].

Non-linear responses between seabird population parameters such as survival and environmental covariates, such as the quadratic relationships modelled in this chap- ter, have previously been suggested. For example, a non-linear effect of prey abun- dance on parameters such as survival and breeding success was predicted by Cairns [1988], who sought to clarify the responses of seabirds to fluctuations in prey. In a later study, Piatt et al. [2007] confirmed that, where there was a significant re- lationship between population parameters and variations in prey density, it was non-linear for several parameters such as fledging and breeding success although relationships with survival were not examined (also in Piatt [1990], Furness and Camphuysen [1997] and Reid et al. [2005]). The survival of Ad´eliepenguins at Ed- monson Point, Antarctica, was found to vary according to a quadratic relationship with winter sea ice extent (a proxy for access to food), suggesting that low and high levels of winter sea ice were detrimental to penguin survival, but there was an inter- mediate optimum level of winter sea ice for survival [Ballerini et al., 2009]. Other techniques to detect non-linear relationships were used by Nevoux and Barbraud [2006] where successive 10-year windows were moved along a time series one year at a time to capture non-linear processes driving population dynamics of white storks within those windows. In that study, authors were able to define a threshold in climatic conditions corresponding to an upper limit above which survival started to be affected [Nevoux and Barbraud, 2006].

In this chapter, I have shown that marine productivity had an age- and spatial- specific association with little penguin survival. In addition, this study linked the physical and chemical processes at an oceanographic level to those which occur at a trophic level (within the food chain, which then affects higher levels through to mesopredators such as penguins). Furthermore, these results have provided insight into how changes to primary productivity may be reflected in seabird survival.

127 It is hoped that the findings in this and the previous chapters will provide the basis for future work; this is discussed in the next chapter.

128 Chapter 8

Summary and future directions

In this final chapter, I present highlights and conclusions of my re- search, and suggest possible directions for future work.

8.1 Summary

The work in this thesis has provided useful evidence on the links between climatic and oceanographic covariates and little penguin survival on Phillip Island Addi- tionally, while temporal variation in survival for first-year and adult little penguins had already been established, this study was the first to investigate potential trends in survival over time, and to link these with the observed status and trends of the population.

Chapter 1

This chapter provided background information about the study and contained my research aims. Here I introduced relevant previous studies on seabirds in general, and discussed where my work fits in with, and advances, previous studies on little penguins. I presented the general and specific aims of my research, and outlined the novelty of my work.

Chapter 2

In this chapter I detailed the organisation of the remarkable 46-year penguin database, a highlight of my research, such that subsets of data based on different penguin at- tributes (such as tag type) could be identified for analysis. I presented the climatic and oceanographic data, and discussed the creation of the covariates used in sub- sequent analyses. This chapter also contained a discussion on the use of Principal Component Analysis as a possible approach to address the high correlations between some covariates. The analysis itself is contained in Appendix D.

129 Chapter 3

This chapter provided details on the notation, methodology and survival models used in my research, including a discussion on banding effects. I illustrated the creation of the sufficient statistics and the likelihood using sample data. I discussed how to determine the goodness of fit of a model and the advantages of the approach used in my research over other available ‘black-box’ type computer packages.

Chapter 4

In this chapter I modelled the time-dependence and banding effect of first-year and adult little penguin survival. I found that first-year birds showed a decreasing trend in survival over time, and there was no statistically significant trend in adult survival. Banded birds had lower survival than transpondered birds. A simple population model confirmed observed trends. A highlight of this chapter was the linking of survival estimates in my study with the observed population status and trends of the little penguins on Phillip Island reported in Sutherland and Dann [2014].

Chapter 5

In this chapter I modelled how adult survival depended on terrestrial covariates (ambient air temperature, rainfall and humidity) during the breeding season and moult period. These are two energetically demanding periods in the lifecycle of a little penguin. Humidity during moult had the most significant effect on adult survival, with a positive association, particularly for banded birds. This was some of the first evidence of the effect of this covariate on little penguin survival. My findings were published in Ganendran et al. [2016].

Chapter 6

In this chapter I modelled the dependence of first-year and adult survival on covari- ates that operate at sea (wind and sea temperature). The survival of first-year birds was most affected by a lagged covariate (east-west wind in the previous spring) with a positive association, particularly for transpondered birds. This result suggested a mechanism that operated via parents since parental fitness before breeding was pre- viously found to be positively correlated with fledging weight, itself correlated with survival in the first twelve months. The survival of adult birds was most affected by an increasing temperature gradient across Bass Strait in the previous winter, par- ticularly for transpondered birds. This chapter highlighted two key factors: first, first-year and adult birds showed different responses to climatic and oceanographic covariates; and second, the effects of climate on little penguin survival may not be immediate but may be lagged.

130 Chapter 7

Several aspects of little penguin foraging behaviour were highlighted in the results from this chapter which modelled how first-year and adult survival depended on marine productivity. I found a quadratic relationship between first-year survival and marine productivity in the Bonney Upwelling area in the summer, possibly suggesting an optimal body mass for first-year survival, and confirming that the westward dispersion of newly fledged birds is a likely adaptive behaviour for survival. This is supported by my finding of a negative relationship between first-year survival and marine productivity in the waters immediately surrounding Phillip Island in the summer (the peak of the breeding season) when competition with breeding adults for food resources would be at its highest.

Adult birds remained close to the colony in summer and winter with positive associ- ations between survival and marine productivity during these times. By remaining close to the colony in summer, adult birds were able to balance the demands of self-maintenance and provisioning chicks during the breeding season, while high marine productivity in waters close to the colony in winter allowed adult birds to build up body mass with positive carry-over effects for the energetic demands of the upcoming breeding season.

In Chapters 5 – 7, I also included the additive and interactive effects of banding with the best climatic and oceanographic covariates on survival.

Appendix B

In a separate smaller study, I investigated if little penguin survival was affected by breeding success. I found that in general, successful breeders have a higher survival probability than unsuccessful breeders. I also investigated the possibility of different effects of climatic and oceanographic covariates on the survival of successful and unsuccessful breeders. For each of the covariates studied, I again found that successful breeders had higher survival than unsuccessful breeders.

Appendix D

I conducted a Principal Component Analysis on all the covariates used in my the- sis. This is an approach sometimes used to reduce large numbers of potentially correlated covariates into a smaller set of ‘synthetic’ covariates (called Principal Components, PCs), each of which is a linear combination of the original covariates. These PCs are then used as covariates in the survival models. In this appendix, I showed that the PCs can be difficult to interpret in a practical sense and impor- tantly, the use of them in survival models did not result in a better fit of the models to the data than the methods used in my thesis.

131 This thesis has been able to advance our understanding of factors which impact little penguin survival. It is hoped that these results can provide a framework for future work into the effect of climate change on penguin survival, and contribute more generally to seabird conservation.

8.2 Future directions

The vast and continually increasing amount of mark-recapture data available through the penguin demography study at Phillip Island allows many other issues relating to penguin survival to be addressed. Here I discuss a few areas in which future work could be considered.

As discussed in Chapter 2, I did not include 659 ‘rebanded’ birds in my analysis, because their records did not reflect their change of tag (either to a new band or to a transponder). However, data on the changeover date were kept. Using these data, the life histories of these birds up to the changeover date could be used in the analysis, and the changeover date treated as the final encounter. Alternatively, the likelihood could be rewritten so that a bird which had a band replaced with a transponder had different survival probabilities for the period that it was banded and the period that it was transpondered. Analysis into the survival of birds that had a change of band would yield valuable insights into the effects of the changeover.

Dann and Cullen [1990] reported that less than 15% of one generation of little penguins produces all the next generation, while Nisbet and Dann [2009] showed that relative productivity of little penguins increased to a peak at about an age of eight years, then declined again. Although a previous study did not find a relationship between survival and reproductive success for little penguins on Phillip Island [Dann et al., 1995], the availability of additional mark-recapture data in the intervening years up to the present time for the same colony and more sophisticated analysis techniques allow the relationship between reproductive timing and success and survival to be further investigated. In this thesis I conducted a study into the effects of breeding success on survival using a small dataset of breeding data from one site on Phillip Island. Future projects using a larger dataset of breeding data for individual birds could investigate if other measures of breeding such as its timing or periods of one or more years not breeding are associated with survival.

While modelling an annual survival provides a useful ‘big picture’, modelling sur- vival for different periods throughout the year may provide useful insights into survival probability during or immediately following annual life-cycle events such as moult and breeding. Johannesen et al. [2002b] found highest survival in the breeding season, and reduced survival in the moult, post-moult and mid-winter

132 periods for little penguins in New Zealand. Tavecchia et al. [2002] and Schaefer et al. [2006] investigated seasonal (summer and winter) and monthly survival of birds, respectively, and Piper [2002] investigated quarterly survival using individ- ual covariates (such as age), group covariates (such as habitat degradation) and universal covariates (such as climate). Analysing the little penguin data to obtain the survival probability for different periods will help to corroborate knowledge of peak mortality periods arising from observed recovery data on dead little penguins washed ashore at different times of the year [Dann et al., 1992].

Dann et al. [2014] found that flipper banding affected the survival of adult little penguins on Phillip Island. Unpublished data currently exist for first-year birds marked with either a flipper band or one of two different sizes of transponder (P Dann, unpublished data). These data can be used to estimate survival probability for banded or transpondered first-year little penguins. In addition, the use of two different sizes of transponder will facilitate the investigation into potential effects of transponder size on first-year survival. Future studies should also consider the inclusion of tag-loss probability for adult birds (as determined by Dann et al. [2014]) into the likelihood equation (Chapter 3).

The use of different analytical tools in the estimation of little penguin survival can be explored in collaboration with experts in other fields of statistics. Bayesian techniques are gaining wider usage in the estimation of population parameters in seabirds (see for example Reid et al. [2013] and Thiebot et al. [2014]), and could be used with the little penguin mark-recapture data for comparison with survival estimates obtained via more traditional approaches.

Each of these suggestions is a large project in its own right.

133

Appendix A

Deriving and programming the likelihood

In Chapter 3, I discussed the derivation of the likelihood using an individual-based approach in which the likelihood contribution of each individual bird was used to construct the likelihood function. Here I present the derivation of the likelihood using a cohort-based approach as in Catchpole et al. [2000]. In a cohort-based approach, the likelihood contribution is taken for each cohort of birds over each mark-recapture occasion, rather than for individual birds. Since this approach is much more computationally efficient, it results in much shorter computer run times, which is particularly desirable given the size of the dataset analysed in this project.

The following is adapted from Catchpole et al. [1998] and Sidhu [2007]. In both of these studies, recovery data were also used, so the parameter λ, the probability of a bird being recovered dead, was included in the likelihood. This parameter is not included here.

A.1 Forming the likelihood

Let t1, . . . , tk denote the k mark-recapture occasions (years), with cohort c (c =

1,...,C) consisting of all birds marked in the same penguin year tc, and C denoting the number of cohorts.

Let

φc,j = Pr (a bird from cohort c, alive at tj, survives until tj+1),

pc,j = Pr (a bird from cohort c, alive at tj+1, is recaptured at tj+1), with j = c, . . . , k−1. j = c corresponds to the initial marking.

Catchpole et al. [1998] defined pc,j as

pc,j = Pr (a bird from cohort c, alive at tj, is recaptured at tj),

135 which is slightly different from the definition of pc,j used here. However, the defini- tion used here does not alter the development of the likelihood.

A.1.1 Sufficient statistics

The sufficient statistics for a cohort-based approach are:

wc,j = number of animals from cohort c recaptured at tj+1, 1 ≤ c ≤ C, c ≤ j ≤ k − 1.

zc,j = number of animals from cohort c not recaptured at tj+1 but seen alive later, 1 ≤ c ≤ C, c ≤ j ≤ k − 1.

vc,j = number of animals from cohort c marked or recaptured at tj and not seen again during the course of the study, 1 ≤ c ≤ C, c ≤ j ≤ k.

This analysis depends on the assumptions that there are no deaths on recapture and that recaptured animals are returned immediately to the population. Note that wc,j refers to recaptures and therefore does not include the initial capture [Catchpole et al., 1998].

The ‘vanishing probability’ χc,j is defined by Catchpole et al. [1998] as

χc,j = Pr (a bird from cohort c, alive at tj, is not seen after tj).

χc,j is a recursive term, and is derived as follows. The following probabilities are conditional on the bird being from cohort c and alive at tj.

χc,j = Pr (not seen after tj)
= Pr dies in (tj, tj+1)

+ Pr (survives till tj+1, not recaptured at tj+1, not seen after tj+1)

= (1 − φc,j) + φc,j(1 − pc,j)χc,j+1, 1 ≤ c ≤ C, c ≤ j ≤ k − 1.

All values of χ can be calculated recursively since

χc,k = Pr (alive in final year of study, not seen thereafter) = 1, 1 ≤ c ≤ C.

A.1.2 The likelihood

I use the same four-bird example as in Chapter 3 (Section 3.3.2, page 43) to show the cohort-based derivation of the likelihood. Assume that these four birds are the only birds from cohort c = 1, that is they were all marked on the same occasion, t1. The individual life histories from six capture-recapture occasions appear below.

136 Recall that a ‘1’ denotes a live capture or recapture and a ‘0’ means that the bird was not encountered on that recapture occasion.

t1 t2 t3 t4 t5 t6 Bird 1 1 1 1 1 0 0 Bird 2 1 0 1 1 0 0 Bird 3 1 0 0 1 0 0 Bird 4 1 1 0 1 0 0

In Chapter 3, I showed that the sufficient statistics for the individual birds were given by       1 1 1 0 0 0 0 0 0 0 0 0 0 1 0 0       0 1 1 0 0 1 0 0 0 0 0 0 0 1 0 0 W =   Z =   V =   . 0 0 1 0 0 1 1 0 0 0 0 0 0 1 0 0       1 0 1 0 0 0 1 0 0 0 0 0 0 1 0 0

Here, since all four birds belong to the first cohort, the cohort-based sufficient statis- tics are obtained by summing the rows of the individual-based sufficient statistics. Thus, the first rows of the cohort-based matrices W , Z and V are now given by:

h i h i h i W1 = 2 2 4 0 0 Z1 = 2 2 0 0 0 V1 = 0 0 0 4 0 0 .

Consider the cohort-based likelihood equation developed by Catchpole et al. [2000]:

C "k−1 k # Y Y wc,j +zc,j wc,j zc,j Y vc,j L = const × φc,j pc,j (1 − pc,j) χc,j . (A.1) c=1 j=c j=c

In the following derivation, I use the cohort-based sufficient statistics to determine the contribution to the liklihood from all the birds in the cohort. I omit the subscript c referring to the cohort to avoid confusion.

All four birds from cohort c = 1 survived to occasion t4. Therefore, their contribu- 4 tion to the survival part of the likelihood is (φ1φ2φ3) .

The first row of W +Z is [4 4 4 0 0]. This represents the total number of birds from this cohort that were recaptured on that occasion, or not recaptured but seen later, that is the number of birds known to be alive on that occasion. Therefore, the survival part of the likelihood is

5 Y wj +zj 4 4 4 0 0 4 φj = φ1φ2φ3φ4φ5 = (φ1φ2φ3) . j=1

137 All four birds were recaptured at occasion t4 and never seen again; therefore the 4 contribution of the ‘vanishing probability’ is χ4. The first row of V is [0 0 0 4 0 0]. This represents the total number of birds that were not seen again after that occasion, that is the number of birds that ‘vanished’ after that occasion. Therefore the ‘vanishing probability’ part of the likelihood is

6 Y vj 4 χj = χ4. j=1

Now consider the recapture part of the likelihood:

2 • two birds were recaptured on occasion t2, with a contribution of p1 2 • two were recaptured on occasion t3, with a contribution of p2 4 • all four birds were recaptured on occasion t4, with a contribution of p3 2 • two birds were not recaptured on occasion t2, with a contribution of (1 − p1) 2 • two birds were not recaptured on occasion t3, with a contribution of (1 − p2) .

Therefore, the contribution to the recapture part of the likelihood is

2 2 4 2 2 p1 p2 p3(1 − p1) (1 − p2) .

Using W1 = [2 2 4 0 0] and Z1 = [2 2 0 0 0], the contribution to the likelihood is

5 Y wj zj 2 2 4 2 2 pj (1 − pj) = p1 p2 p3(1 − p1) (1 − p2) . j=1

The product of each of these terms represents the likelihood contribution from the first cohort. Once the likelihood contributions from each cohort are determined, these are multiplied together as in Equation A.1.

A.2 Model fitting

In my work, I consider two subsets of the raw data: banded birds and transpondered birds. Each of these subsets contains birds marked as chicks and birds marked as adults of unknown age. For ease of calculation, I set up cohort-based sufficient statistics for each of these subsets, that is V , W and Z for banded birds and for transpondered birds. The matrices for transpondered birds (with fewer years of data) are padded with zeros so that they are equal in size to those of the banded birds.

138 The first year for the life-history data is 1967 (that is the breeding season or Penguin Year 1967/68), and the first year of birth is set at 1963 in order to incorporate adult birds of unknown age in the first cohort. If the first year of birth is set to be the same as the first year of history, then only chicks that hatched that year (and no adults) would have been included in that cohort. The last year of the life-history data is 2012, and the last year of birth is set at 2011 in order to incorporate birds hatched in the 2011/12 breeding season.

When considering covariate dependence for a parameter or for an age class of a parameter, I define appropriate covariate matrices. For example, the covariate matrix for the survival probability of the first age class (that is, first-year birds) may have been a matrix with two columns – Column 1: mean rainfall in a particular season; Column 2: mean ambient temperature in a particular season.

The following example illustrates how the number of parameters in a model are cal- culated. Consider a model with the survival age structure of (1,2) and dependence (C,T). This is interpreted as first-year birds having constant survival probability, and adult birds having time-dependent survival probability. Since the first age com- ponent, φ1, is constant, there is one parameter for this age group. The second age component, φ2, is time-dependent, so the number of parameters for this age group is the number of recapture occasions minus 1 (that is, 46 − 1 = 45). Hence, for this model, the total number of survival parameters is 46. I write this survival model as

φ1(C), φ2(t).

The same procedure applies in determining the number of parameters for the recap- ture probability. For example, a recapture age structure of (1,1,3) indicates separate recapture probabilities in the first year of life, the second year of life, and the third and subsequent years of life. Additionally, a dependence of (C,C,C) indicates that the recapture probabilities for each of these age classes is constant. Thus, there is one parameter for each of the age classes, giving a total number of recapture parameters of 3.

In my work I used an age structure for recapture probability of (1,1,3) and an age + time dependence for the recapture probability. This means that the recap- ture probability is allowed to be fully time-dependent, with the same time variation for each age class, differing only by a constant value. For example, the recapture probability of birds in their third and subsequent years of life is equal to the recap- ture probability of first-year birds in that year plus a constant (on a logit scale). Thus, there are 45 parameters for the time-varying dependence for one age class as well as two constants separating the three age classes, giving a total of 47 parameters for recapture.

139 For a model defined as φ1(C), φ2(T )/p1, p2, p3(age + time), the total number of pa- rameters is 93, made up of 46 survival parameters and 47 recapture parameters.

In the MATLAB programs that I used, once the model is defined, the appropriate form of the parameter matrix corresponding to the model structure for that pa- rameter is determined. To illustrate this, consider an example of one age class of birds with time-dependent survival. The parameter matrix for φ has 45 columns (number of recapture occasions minus 1) and 49 rows (corresponding to the number of cohorts, that is, from the first year of birth 1963, as explained above, to 2011).

Recall that the logit link function ensures that the parameter estimates lie between 0 and 1. I use θ to denote the survival parameters on a logistic scale. Thus θ = logit (φ) and φ = ilogit (θ) (abbreviated as ilt (θ)).

The survival for birds from cohort c from capture-recapture occasion tj to tj+1 th th (i.e. the j year of the study), φc,j, is given by the (c, j) entry of the parameter matrix, so that φ2,4 = ilogit (θ2,4) is the survival of birds from the second cohort in the fourth year of the study (i.e. the 1971 calendar year). The parameter matrix for φ is of the form:

  ilt(θ1,1) ilt(θ1,2) ilt(θ1,3) ilt(θ1,4) ··· ilt(θ1,45)    ilt(θ2,2) ilt(θ2,3) ilt(θ2,4) ··· ilt(θ2,45)     ilt(θ ) ilt(θ ) ··· ilt(θ )   3,3 3,4 3,45   . . .  .  .. .. .       ilt(θ48,44) ilt(θ48,45)  ilt(θ49,45)

I used several functions and programs in MATLAB to fit a model, adapted from the original code written by E. Catchpole and used in Catchpole et al. [2000]. All code is available from [email protected].

140 Appendix B

Survival, breeding success and significant climatic and oceanographic covariates

B.1 Introduction

The trade-off between reproduction and survival suggests that current reproductive output should be negatively correlated with survival [Stearns, 1992], that is, the ‘cost’ or ‘penalty’ incurred by reproducing is measured by a decrease in survival and future reproduction [Weimerskirch, 1992, Sæther et al., 1993, Zera and Harsh- man, 2001]. However, studies instead suggest that the ‘quality’ of an individual is positively correlated with survival, and quality is often indicated by its reproductive effort. For example, Lescro¨elet al. [2009] reported that successful breeders exhib- ited higher survival than deferred and unsuccessful breeders, and found a quadratic relationship between apparent survival and a breeding-quality index. Le Vaillant et al. [2015] used an index of breeding quality to relate foraging ability (a main con- tributor to survival) of king penguins and their breeding, finding that birds of better breeding quality performed shorter foraging trips at sea during the first period of the cr`eching phase.

Studies carried out on little penguin populations have also found positive relation- ships between breeding success and survival, with possible confounding effects of age and experience. Dann and Cullen [1990] found that lifetime reproductive out- put was positively related with age for both male and female little penguins, and Johannesen et al. [2003] found evidence that the probability of laying two clutches increased with age. That study also determined that little penguins that laid two clutches in a breeding season had a higher probability of laying two clutches in the following breeding season and a higher survival probability; this was interpreted as a difference in quality between the birds.

However, differences between individuals in a population mean that some birds outperform others in breeding success; these differences may be due to individual

141 heterogeneity in quality or to age- and experience-related processes [Lescro¨elet al., 2009, Nisbet and Dann, 2009] and survival [Dann and Cullen, 1990]. These out- performing individuals may also exhibit traits such as better foraging efficiency — a main contributor to successfully raising offspring and to surviving the following year [Jenouvrier et al., 2015]. A useful review of reproductive success in relation to the ‘quality’ of a bird and the timing of breeding is contained in Verhulst and Nilsson [2008].

To test the hypothesis that higher reproductive performance is positively related to higher survival for little penguins, I first analysed a subset of breeding data from one study area on Phillip Island. I identified ‘unsuccessful breeders’ as those with no chicks fledged over the time period examined and ‘successful breeders’ as those with at least one chick fledged over the time period examined, and modelled the relationship between breeding success and survival. I then examined the effect of potential interactions between breeding and the most significant climatic and oceanographic covariates from Chapters 5 – 7 on survival.

B.2 Breeding data

An important first step was to meaningfully summarise the breeding data. Breed- ing data have been collected since 1968, with more sites added each year from 1982 [Nisbet and Dann, 2009]. For this analysis, I used breeding data from Location 29, Site 29. This is the site with the largest number of data points for each year from 1986 (first year of breeding data from this site) to 2012 and, importantly, it is also one of the sites from which data were used for the main analysis conducted in this thesis. I compared the mean number of chicks fledged per pair using this subset of data with that from all sites on Phillip Island. Figure B.1 shows a very strong correlation between these two sets of values (r = 0.95, p < 10−2), which pro- vides confidence that this subset of breeding data is representative of the breeding population from all sites on Phillip Island.

I next created a breeding covariate to use in survival models, and limited the breed- ing data from this site to birds transpondered as adults from 2001 (first year of history for birds in the breeding subset) to 2012 for two main reasons. Firstly, this time frame most closely matched that of the covariate (1999 to 2012) used in Chapter 7, and is beyond the problematic transition period from flipper banding to transponders from the late 1990s to mid-2000s (Table 2.8). Secondly, I only used adult birds because the commencement of breeding for a bird tagged as a chick is not always obvious whereas a bird tagged as an adult is more likely to be a breeding

142 bird. Furthermore, the small chick sample size (n = 53) would result in large stan- dard errors. The final subset of breeding data used in the study in this appendix contained 453 transpondered adult birds.

1.8 all sites Location 29, Site 29 1.6

1.4

1.2

1

0.8

0.6

0.4 Mean number of chicks fledged per pair

0.2

0 1990 1995 2000 2005 2010 2015 Year

Figure B.1: Mean numbers of chicks fledged per pair from all sites on Phillip Island (solid line) and from Location 29, Site 29 (dashed line). (These data include birds tagged as chicks or as adults of unknown age.)

Breeding data over an individual’s lifetime may also be incomplete in some instances as a previously seen bird may continue to breed beyond the end of the study. To accommodate this, Dann and Cullen [1990] did not use the last three years of their data, concluding that a bird not seen for that number of years is likely to have died.

A bird tagged as an adult at one site may have begun breeding at a different site. Any interruption to breeding or an end to the breeding history of a bird may be due to the bird temporarily or permanently moving out of the study area. If a bird was seen but had no breeding data, it may have returned to the colony to moult and not to breed. The majority of birds tagged as chicks were never seen again after the initial tagging. Sidhu et al. [2007] reported that 87% of birds flipper banded as chicks were not seen again after the initial tagging. Using the full dataset from my thesis, I calculated this number to be 75%.

143 B.3 Breeding data and recapture histories

An interesting aspect of this study was the comparison of breeding and life history data for the same individual, providing useful insights into breeding patterns. I extracted the recapture history for each bird in the breeding dataset from Location 29, Site 29. A sample of data showing the recapture history (first row of data) and breeding history (second row) for two birds viewed together is shown below. The column names are: ‘id’ - transponder number; ‘h01’ to ‘h12’ are the years 2001 to 2012. In the first row for each bird, ‘1’ indicates that the bird was recaptured on that occasion, and ‘0’ indicates it was not recaptured. In the second row ‘X’ indicates the period before the bird’s first encounter, ‘-’ indicates that no breeding data were recorded (no eggs or chicks recorded), and the numbers are the total number of chicks fledged by that bird in that year. A ‘0’ in the breeding data indicates that the bird attempted breeding but was unsuccessful, i.e. no chicks were fledged.

id h01 h02 h03 h04 h05 h06 h07 h08 h09 h10 h11 h12 652B2D7 0 0 1 1 1 1 1 1 1 1 0 0 652B2D7 X X - - - 2 - - 4 - - -

652BBC6 0 0 1 0 1 0 1 1 1 1 1 1 652BBC6 X X - - 1 - - 0 - - - -

The data for the first bird (ID 652B2D7) illustrate the difficulty in determining whether or not a bird has completed its lifetime breeding output. It was first encountered and tagged as an adult of unknown age in 2003 (but did not breed), and recaptured in 2004 and 2005 with no breeding data recorded at these times. It began breeding at this site in 2006, with two fledged chicks. It was seen but not recorded breeding for two years, then bred successfully in 2009 with four fledged chicks. It was seen the following year (but was not recorded breeding) and was not seen in 2011 or 2012. It may have moved outside the study area to breed or it may have died.

The second bird (ID 652BBC6) shows a similar pattern of not breeding on the first occasion that it was captured (2003), not being seen the following year, having a successful attempt at breeding (2005), not being recorded breeding for two years, followed by an unsuccessful breeding attempt in 2008. No further breeding data were recorded for this bird in the following four years, even though it was seen during that time.

144 B.4 Some breeding statistics

Breeding performance varies from year to year [Reilly and Cullen, 1981], with Dann and Cullen [1990] reporting a mean figure of 0.84 chicks fledged per pair, ranging from a low of 0.17 to a high of 1.82 using data from 1968 to 1987. My analysis of the breeding data from Location 29, Site 29 over the period 1986 to 2012 indicated a mean figure of 1.08, ranging from a low of 0.09 in 1997 to a high of 1.79 in 1990. For the 453 adult birds in my study, the mean time span from the first to last encounter of a bird (lifetime) was 5.23 years, the mean number of breeding attempts recorded was 1.66 attempts per bird per lifetime, the mean number of successful breeding attempts was 0.95 per bird and the mean number of chicks fledged was 1.54 per bird.

A summary of breeding output (including the number of fledged chicks per breeding attempt per year) of birds seen each year is given in Table B.1. The fourth column is the proportion of total birds seen that year that attempted to breed (i.e. eggs were seen). The fifth column is the number of attempts from which no chicks were fledged; in the next column, the number of attempts from which one chick was fledged, and so on. The last column is the total number of chicks fledged that year.

Table B.1: Number of breeding attempts, the proportion of birds attempting breed- ing, the number of chicks fledged per breeding attempt and the total number of chicks fledged each year. Data from Location 29, Site 29.

ber of attemptsortion attempting ear otal birds seen chicks chick chickschickschicksotal fledged Y T Num Prop 0 1 2 3 4 T

2001 11 0 0.00 0 0 0 0 0 0 2002 66 36 0.55 21 9 4 2 0 15 2003 114 63 0.55 28 15 20 0 0 35 2004 146 66 0.45 38 21 5 0 0 26 2005 192 84 0.44 25 16 41 2 0 59 2006 214 74 0.35 37 13 24 0 0 37 2007 240 80 0.33 33 16 31 0 0 47 2008 221 72 0.33 35 25 12 0 0 37 2009 240 85 0.35 34 18 27 3 3 51 2010 242 64 0.26 22 13 29 0 0 42 2011 236 60 0.25 15 19 25 1 0 45 2012 226 66 0.29 28 16 21 1 0 38

145 B.5 Measuring breeding effects

Mark-recapture and breeding data can be extracted from the main dataset for individual birds, making it possible to explore a bird’s survival in relation to its breeding output. However, it should be noted that this analysis is based on far fewer birds and a much shorter time period than the main analyses conducted in the body of this thesis. As a first step, I separated the ‘successful’ breeders (at least one chick fledged, n = 299) from the ‘unsuccessful’ breeders (no chicks fledged in their lifetime, n = 135) and modelled time-dependent survival for each group (Figure B.2). The lines indicate a decreasing trend of survival with time; these are statistically significant (p < 10−2 for both). There was a sharp increase in the survival of unsuccessful breeders in 2005, possibly an indication of these birds’ energetic efforts in survival ahead of breeding [Ropert-Coudert et al., 2004b, Jenouvrier et al., 2005b]. Note that although 11 birds were seen in 2001, there were no breeding attempts (i.e. no successful breeders, see Table B.1), therefore the first survival estimate shown in Figure B.2 is for 2003 (as there were no successful breeders ‘recaptured’ in 2002). (Refer to Appendix A for a full explanation of the formation of the likelihood.)

1 successful breeder

0.9

0.8

0.7 unsuccessful breeder Survival

0.6

0.5

0.4 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Year

Figure B.2: Survival over time of successful (solid curve) and unsuccessful (dashed curve) breeders. The lines show the linear relationship between survival and time, which is used as a covariate in the models.

146 B.6 Breeding patterns and climatic and oceanographic covariates — results and discussion

In this study, I have examined whether or not breeding success (i.e. whether a bird was a successful/unsuccessful breeder) and the most significant climate covariates from the survival models had an interactive effect on survival. I used the covari- ates that had the biggest effect on survival (those which accounted for the highest proportion of deviance) in the models in Chapters 5 to 7.

Breeding patterns and climatic and oceanographic covariates

My analysis has shown that successful breeders in general have higher survival than unsuccessful breeders (Figure B.2). When climate covariates are included in survival models, this is still the case (Figures B.3, B.4 and B.5). Each of the survival models shown in these figures is paired with a histogram of the covariate used in that model. In Figure B.3, successful breeders show a statistically significant increase in survival in response to the covariate (mean humidity during moult), and unsuccessful breeders show a statistically significant decrease in survival (p < 10−2 for both). In Figure B.4, successful breeders show a statistically significant increase in survival in response to the covariate (residuals of mean east-west sea temperature gradient in the previous winter, p < 10−2) while unsuccessful breeders show no trend (p = 0.9). In Figure B.5, both successful and unsuccessful breeders show no trend in survival in response to the covariate (mean chlorophyll a concentration in the waters surrounding Phillip Island, p = 0.3 and 0.9 respectively).

If breeding success is an indicator of the quality of a bird in terms of its fitness traits, then it is not surprising that successful breeders have higher survival than unsuc- cessful breeders, and in some instances as shown in this study, successful breeders survive better under the effects of some climate covariates. The negative or ab- sence of effects of covariates on unsuccessful breeders was perhaps more indicative of confounding effects such as lower body condition, and less of a negative associa- tion with the covariate. For example, in Figure B.3, the covariate (mean humidity during moult) had a positive effect on successful breeders and a negative effect on unsuccessful breeders, although a previous study has shown that this covariate has a positive association with survival [Ganendran et al., 2016]. Body condition has previously been linked to breeding performance of seabirds. Poor body condition in Magellanic penguins caused by depletion of fat reserves was a cause of nest deser- tion at Punta Tumbo [Frere et al., 1998]. Heavier Ad´eliepenguins with higher fat stores to sustain long fasts have higher reproductive success than lighter penguins, and birds in poor conditions do not attempt to breed [Vleck and Vleck, 2002]. Male Ad´eliepenguins foraging closer to the colony at B´echervaise Island are unable to

147 maintain their own body reserves and must make longer trips, particularly during the pre-hatch and early guard periods [Clarke, 2001]. Male eastern rockhopper penguins Eudyptes chrysocome filholi, having fasted through the first 3 to 4 weeks of the chick guard phase, must spend more time feeding themselves rather than contributing to provisioning chicks in the later stages of chick guarding [Morrison et al., 2016].

The link between body condition and breeding success has been reported in lit- tle penguins. Increased egg desertion by little penguins in Motuara Island, New Zealand was significantly correlated with poorer body condition and longer forag- ing trips [Numata et al., 2000]. Poor body condition in little penguins on Phillip Island during the 31 – 40 day incubation period (when parents take it in turn to guard the eggs, [Reilly and Cullen, 1981]) has been correlated with lower breeding success, with birds of lower body condition less likely to rear chicks successfully [Robinson et al., 2005]. A carry-over effect from winter to breeding was found for female and male little penguins; those with higher body mass in winter were more likely to breed early, with the males more likely to breed successfully [Salton et al., 2015].

The differences in the effects of climatic and oceanographic covariates on breeding success may have other confounding factors such as laying date; studies show that earlier breeding is associated with greater breeding success. Almost all late breeding king penguins on Possession Island were unsuccessful and only half of the successful breeders bred the following season; those that did bred late in the season [Weimer- skirch et al., 1992]. Late-breeding chinstrap penguins on Deception Island were in poorer health than early breeders, and were more likely to experience breeding fail- ure [Moreno et al., 1998]. Humboldt penguins at Punta San Juan, Peru, laying two clutches earlier in the season had higher breeding success than those laying later [Paredes et al., 2002], and late-laying little penguins on Phillip Island were more likely to experience hatching failure than early-laying birds [Nisbet and Dann, 2009].

The mechanisms by which climate influences breeding productivity, and ultimately population changes, are varied. For example, differences in sea-ice extent at a colony of emperor penguins in Terre Ad´eliehad contrasting effects on population dynamics; an increase in sea-ice in winter was negatively correlated with hatching success but positively correlated with survival [Barbraud and Weimerskirch, 2001]. El Ni˜noand La Ni˜naperturbations were correlated with the breeding success of little penguins in New Zealand, with later breeding and a lower proportion of double breeders observed during La Ni˜naconditions [Perriman et al., 2000]. The timing and breeding success of little penguins on Phillip Island were associated with changes in sea-surface temperatures, with higher temperatures associated with early laying,

148 higher chick mass at fledging and higher numbers of chicks produced per breeding pair [Cullen et al., 2009].

In this study, I investigated relationships between breeding success, climate and survival of little penguins. I found that successful breeders had higher survival than unsuccessful breeders. In addition, successful and unsuccessful breeders responded differently to climate covariates. Successful breeders showed increasing survival in response to the covariates in Figures B.3 and B.4. Breeding success was not a statistically significant factor in the association between the mean chlorophyll a concentration in waters surrounding Phillip Island in the summer and survival (Figure B.5).

This study was limited by the small sample size of 453 adult birds of unknown age, and I was not able to test for the confounding effects of age and experience. It serves as a useful preliminary investigation into the interactive effects of breeding, climatic and oceanographic covariates, and the survival of little penguins on Phillip Island. It has provided evidence for the positive association between breeding suc- cess and survival, and for the differing effects of climate on the survival of successful and unsuccessful breeders. Clearly further study with a larger number of birds is warranted and a useful objective would be to incorporate age and breeding success into future survival models.

149 1

successful breeder 0.95

0.9

Survival unsuccessful breeder 0.85

0.8

0.75 12.5 13 13.5 14 14.5 15 Mean humidity during moult (hPa)

10

9

8

7

6

5

4

3

2

1

0 12 12.5 13 13.5 14 14.5 15 15.5 Mean humidity during moult (hPa) Figure B.3: Top: Interactive effect of breeding success and mean humidity during moult on survival. Survival of successful breeders is shown with a solid line and unsuccessful breeders with a dashed line. Bottom: The distribution of the mean humidity during the mould period over the length of the study (1968 – 2012).

150 0.96

0.94

0.92 successful breeder 0.9

Survival 0.88

0.86 unsuccessful breeder

0.84

0.82 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 Residuals of mean east−west sea temperature gradient in previous winter

12

10

8

6

4

2

0 −0.8 −0.6 −0.4 −0.2 0 0.2 0.4 0.6 0.8 Residuals of mean east−west sea temperature gradient in previous winter Figure B.4: Top: Interactive effect of breeding success and residuals of mean east- west sea temperature gradient in the previous winter on survival. Survival of suc- cessful breeders is shown with a solid line and unsuccessful breeders with a dashed line. Bottom: The distribution of the residuals of mean east-west sea temperature gradient in the previous winter (1968 – 2012).

151 1

0.98

0.96 successful breeder

0.94

0.92

Survival 0.9

0.88

unsuccessful breeder 0.86

0.84

0.82 1 1.2 1.4 1.6 1.8 2 2.2 2.4 2.6 2.8 Mean chlorophyll a concentration in Box 1 in current summer (mg/m3)

5

4.5

4

3.5

3

2.5

2

1.5

1

0.5

0 0.5 1 1.5 2 2.5 3 Mean chlorophyll a concentration in Box 1 in current summer (mg/m3) Figure B.5: Top: Interactive effect of breeding success and the mean chlorophyll a concentration in the waters immediately surrounding Phillip Island in the current summer (Box 1) on survival. Survival of successful breeders is shown with a solid line and unsuccessful breeders with a dashed line. Bottom: The distribution of the mean chlorophyll a concentration in the waters immediately surrounding Phillip Island (Box 1) in the current summer.

152 Appendix C

Covariate and correlation matrices

C.1 Air temperature and rainfall

Table C.1: Air temperature and rainfall covariates.

Label Description RB Total rainfall in breeding season RM Total rainfall in moult period TB Mean ambient air temperature in breeding season TM Mean ambient air temperature in moult period D27B Number of degree days over 27◦ in breeding season D27M Number of degree days over 27◦ in moult period D35B Number of degree days over 35◦ in breeding season D35M Number of degree days over 35◦ in moult period VB Mean vapour pressure in breeding season VM Mean vapour pressure in moult period

153 20 80 21 23 0 30 60 0 4 12.0 14.5

●● ●● ● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ●● ●● ●● ● ● ● ● ● ●●●●●● ●●●●●●●● ●●●●●●●● ●●●●●●● ●●●●● ● ●●● ● ● ● ● ●●● ●● ● ●● ●●● ● ● ●●●●● ● ●● ●●● ● ● ● ●●● ● ● ●● ●● ●● ● ●●●●● ●●●● ● ● ● ●●● ●●●● ●● ●●● ● ● ●●●●●● ● ● ● ●●●● ●● ● ●●●●●● ● ●● ●●●● ●●●●● ●●● ●● ● ● ●● ●●●●● ●●● ●● ● ● RB ●●●●●●●● ● ●●●●●●●●●●● ●● ●●●●● ● ●●●●● ●●●●● ●●●●●●● ● ●●●● ● ● ● ●● ● ●●●●●●●● ● ●●●●●●●●● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ● ●● ● ●● ●●●● ● ● ● ●● ● ●●● ●● ●● 30 80 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●●● ● ●● ● ● ●●● ● ●● ●●● ● ● ● ●● ● ● ● ●● ●●●● ●● ● ●●●●●● ●● ●● ● ●●●●●●●● ●● ●●●●●●● ●● ● ● ●●●●●● ●● ●● ●●● ● ●●●● ●●● ●●●●●● ●● ● ●● ●● ●●● ● ●●● ●●●● ● RM ● ●●●●● ● ●●●● ●● ● ●●●●●●●●● ● ●●●●●●●● ●● ●● ● ● ● ●● ●●●●●●● ● ● ●●●●●●●●● ●● ● ●●●●●● ● ●●●●●● ●●●●●● ● ●●●●● ●●● ●●●●● ● ●●● ●● ● ● ● ● ●●●●● ● ●●●●● ●●●● ● ●●●● ● ●● ● ● ●● ● ● ●●●● ●●●● ● ● ●●● ● ●●●● ● ● ● ●● 20 100 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●●●●●●● ●●●●●●●●● ● ● ●●●●●●● ●● ●●●●●●●●● ●●●● ●●●● ● ●●●●●● ● ● ● ● ● ●●● ●●●●●●●● ● ●● ●●●● ●● ●● ● ● ● ●●● ●●●● ●● ● ● ●● ● ● ● ● ● ● ●●● ●● ●●●●● ● ●●●●● ● ●●●●●●● ● ●●●●●●● ●●●●●●● ● ●●●●● ●● ●● ● ●●●●● ●● ●●●● ●●● ● ● ●●●● ●● ●● ●●● ● ●●●● ● ●●●● ●●●● ● ●●● ● ● ●●● ● ●● ●● ● ● ●●● ● ●● ●●●●●● TB ●●●● ●● ●●●●● ●●●●● ● ●●●● ● ● ●●● ●● ● ●●● ●●● ● ●● ● ● ● ●● ●● ●● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● 19.0 ● ● ● ● ● ● ● ● ● ● ●● ● ●●●● ● ●●● ●● ●● ● ●● ● ● ● ● ● ● ● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ● ● ●● ●●●●●● ● ●●● ● ● ● ●●● ● ●● ●●● ● ●●● ●● ●●●●● ● ● ● ● ●●● ● ● ●●●●●●● ● ●●●● ●● ● ●●●●●●● ● ●● ●●●●● ● ●●● ●●● ●●●●●● ● ●●● ● ● ●● ● ● ●● ●● ●● ●●● ●●●● ●●●●●●● ●●●● ● ● ●●●●●● ● ●● ●●● ● ●●●●● ●●●● ● ● ●● ●●● ● ●● ● ● ●●●●●● ●● ●●●●● ● ●●●●●● ● TM ●●●●● ●● ●●●●●●● ●● ● ● ●● ●●●●●●● ●●●●●●● ●● ● ● ●●● ●●●● ●●● ● ●● ● ●●●● ● ● ●● ● ● ● ●● 21 24 ● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ●

●●●●●● ●●●● ● ● ●●● ●●●● ● ● ●●● ● ● ●● ● ● ● ●●●●● ● ● ●● ● ●● ●●●● ● ●●●●●● ● ●●●●●● ● ●● ●● ●● ●●●●●● ●●●● ● ● ● ● ●●● ●●● ●● ●● ●●●● ●●●●● ●● ●● ● ●●●● ●●● ●●● ●●●●● ● ● ●● ● ● ● ●● ● ● ●●● ●●●● ● ●●●● ●●● ●●●●●●●● ●●●●● ● ●● D27B ●●●●●●● ●●● ● ●● ●●●●●●●●● ●● ●●● ● ● ●● ● ●●● ●●●●●● ●●●●●● ●●●●●● ●●●●● ●● ●●●●● ●● ● ●● ●●● ●● ●● ●●●● ●●●●●● ●●●●● ● ●●●●●● ● ●●● ● ●●● ● ●● ● ●●●●●●● ●● ●●● ●● 20 120

● ● ● ● ● ● ● ● ●

● ●●● ● ●●●●●●● ●●● ●●●●● ● ●●●●● ● ●●●●● ●●● ●● ● ● ● ●● ● ●●●●● ● ● ● ●●●● ●● ●● ●●●● ●●●●● ●● ● ●● ●● ● ● ●● ●● ●● ●● ●● ● ● ●● ● ● ●●● ● ● ●●●●●●●● ●●●●●● ● ●●●●●●● ● ●●●●●● ●● ●●●● ● ●●● D27M ●● ●● ● ● ●●●●●●●● ● ●● ●● ● ●●●●●●● ● ●●●●● ● ● ●●●●●●● ● ● ●●●●●● ●● ●●● ● ●●● ● ● ● ●● ●● ● ● ●● ●●●●● ● ● ●●●●●● ●●● ● ●● ●●● ● ●●●●● ● ●●●● ● ●●●●● ● ●●●●● ● ● ● ●● ●●● ●● ●● ●●●●● 0 40

● ● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ● ●●●●●●●● ●●●●● ●●●● ●● ●● ● ●●●● ●●●●●● ● ● ● ● ● ●●●●● ● ●●●●●● ● ● ● ● ● ●● ● ● ●● ● ● ●● ● ● ●● ● ●● D35B ● ●●● ● ●● ●● ●●●●● ● ●● ●● ● ●●● ●● ●●●● ●● ●●●●● ●●● ● ● ●● ●●● ●●● ● ● ● ●●●●●●●● ●●● ●●● ●●●●●●●● ●●●● ●●● ●●●●●●●● ● ●●●●●●●●● ● ●●●●●●●● ●●●●● ● ●● ● ●●●●●●●●●● ●● ●● ●● ●●●●●●●●●●●●●● ●● ●●●●●●● ●●● ● ●●●●● ●●●●●●● ●●●● ●●●● ●●●●●●● ●● ●●● ● 0 15

● ● ● ● ● ● ● ● ●

● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ●● ● ● ● ● ● D35M ● ●● ● ● ● ●●●● ●● ●●●●●●●●●●●● ●●●●●● ●● ●●● ● ●● ●●●●● ●●●●●●●●●●●●●●●●●●●●●●●●●●●●●● ● ●●● ●●● ● ● ● ●●●●●●●●● ●● ●● ●●●●●● ●●●●●● ●●● ●● ●●●●●●●●●●●● ●●●●● ●● ●●● ●●● ● ●●●●●● ●●● ●●●●●●●●●●●●●●●●●● ●● ●●●●●●●●●●●●● ●● ● ● ●● ●●●●●●●●●●● ●●●●●● ●●●●●●●● ● ●●●●●● ●● ●●●● ●●●●●●●●● ●●●●●●● ●●●●●●●●●● ●● ● ● ●●●● ●● ● ●●●●●●●●●●●●●●●● ●●●●●●● ●●●●● ●●●●● ● ●● ●● ●●●●●●●●●● ● ●●● ●●●● ● ●● ●●●●●● ●●● ●●● ● ●●●●● 0 4

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ●● ● ●● ●●●● ●● ● ●● ●●● ● ● ●●● ●● ●●● ● ● ●●●●● ●● ●● ● ● ●●●● ● ●● ●●●● ● ●●●● ● ● ● ●●●●●● ●●● ●● ● ●●●● ● ●●●● ●● ●● ● ● ● ● ●● ● ●● ●●●●● ● ● ●●●●● ● ●●●●●●●● ●●●●●●● ● ●●●●● ●● ●●●●●●● ● ●● ●● ● ● ● ● ●●●●●●● ● ● ●●●●● ●●●● ● ● ●●● ● ●●●● ● ●●●●●●● ●●●● ●● ●● ● ● ● VB ●●●● ●● ● ●●● ●●● ● ●●●● ● ● ● ●● ● ● ●● ●● ● ●● ● ●●● ● ●●●●● ●● ●●● ● ● ● ●● ●●●● ● ● ● ●● ● ● ● ●● ●● ●● ● ● ● ●● ● ● 11.0 13.5 ●● ●● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ●●● ● ●● ●●● ● ● ● ● ● ● ● ● ●● ● ●● ●● ● ●● ● ● ● ● ●●● ●● ● ●● ● ●●●● ●●● ●● ●● ● ● ● ●● ●●● ● ● ●● ●●● ●●●●● ●● ●●●●●●● ● ●●● ●●● ● ●●●●●● ● ●●●●● ● ●●● ● ● ● ● ● ●●●● ● ●●●●● ●●●●●● ●●● ●●● ● ●●●●● ●●●●●● ●● ● ●●●● ● ●● ● ● ● ●● ● ●●●●●●●● ● ●●● ● ●●● ● ● ●● ●●● ●●●●● ● ●● ●● ●● ●● ●● ● ● ● ●● ●●● VM ●●● ● ● ●● ●● ● ● ● ● ● ●● ● ● ●● ● ● ● ●● ● ●●● ● ● ● ●●● ● ●●●● ●●● ● ●●●●● ●●●●● ●●●●●● ●●●●● ●● ● ● ●● ● ●●● 12.0 30 70 19.0 21.5 20 100 0 10 20 11.0 13.0

Figure C.1: Scatterplot matrix for air temperature and rainfall covariates

154 RB RM TB TM D27B D27M D35B D35M VB VM 1

RB ● ● ● ● ● ● 0.8 RM ● ● ● ● ● ● ● ● ● 0.6 TB ● ● ● ● 0.4 TM ● ● ● ● 0.2 D27B ● ● ● ● ● ● 0

D27M ● ● ● ● ● ● −0.2

D35B ● ● ● ● ● ● ● −0.4 D35M ● ● ● ● ● ● ● ● −0.6 VB ● ● ● ● ● ● −0.8

VM ● ● ● ● ● ● −1 Figure C.2: Correlation matrix for air temperature and rainfall covariates. The areas of the circles show the absolute value of the corresponding correlation coeffi- cients.

155 Table C.2: Correlation coefficients for air temperature and rainfall covariates.

RB RM TB TM D27B D27M D35B D35M VB VM RB 1 0.14 -0.28 -0.34 -0.25 -0.53 -0.26 -0.14 0.53 0 RM 1 0.04 -0.21 -0.03 -0.24 -0.09 -0.1 0.13 0.15 TB 1 0.6 0.83 0.24 0.59 0.15 0.37 0.34 TM 1 0.35 0.33 0.22 -0.06 0.2 0.57 D27B 1 0.16 0.75 0.16 0.25 0.23 D27M 1 0 0.66 -0.17 -0.02 D35B 1 0.05 0.02 -0.03 D35B 1 -0.04 -0.23 VB 1 0.42 VM 1

Table C.3: p-values for correlation coefficients for air temperature and rainfall co- variates.

RB RM TB TM D27B D27M D35B D35M VB VM RB 1 0.36 0.06 0.02 0.1 0 0.08 0.35 0 0.99 RM 1 0.82 0.17 0.86 0.11 0.58 0.53 0.38 0.34 TB 1 0 0 0.11 0 0.31 0.01 0.02 TM 1 0.02 0.02 0.15 0.68 0.18 0 D27B 1 0.31 0 0.3 0.1 0.14 D27M 1 1 0 0.27 0.9 D35B 1 0.73 0.91 0.84 D35B 1 0.77 0.13 VB 1 0 VM 1

156 C.2 Wind and sea temperature

Table C.4: Wind and sea temperature covariates.

Covariate label Description WS Mean wind speed WD Strong-wind days EWW East-west wind SST Sea-surface temperature TGR Temperature gradient WiP Previous winter SpP Previous spring SuC Current summer AuC Current autumn WiC Current winter SpC Current spring

157 EWW_WiP EWW_SpP EWW_SuC EWW_AuC EWW_WiC EWW_WSpC WS_WiP WS_SpP WS_SuC WS_AuC WS_WiC WS_WSpC WD_WiP WD_SpP WD_SuC WD_AuC WD_WiC WD_SpC SST_WiP SST_SpP SST_SuC SST_AuC SST_WiC SST_SpC TGR_WiP TGR_SpP TGR_SuC TGR_AuC TGR_WiC TGR_SpC EWW_WiP 1 EWW_SpP EWW_SuC EWW_AuC 0.8 EWW_WiC EWW_WSpC WS_WiP 0.6 WS_SpP WS_SuC WS_AuC 0.4 WS_WiC WS_WSpC WD_WiP 0.2 WD_SpP WD_SuC WD_AuC 0 WD_WiC WD_SpC SST_WiP −0.2 SST_SpP SST_SuC SST_AuC −0.4 SST_WiC SST_SpC TGR_WiP −0.6 TGR_SpP TGR_SuC TGR_AuC −0.8 TGR_WiC TGR_SpC −1

Figure C.3: Correlation matrix for wind and sea temperature covariates. The areas of the circles show the absolute value of the corresponding correlation coefficients.

158 Table C.5: Correlation coefficients for wind and sea temperature covariates. WiP WiC SpP SuC AuC SpC WiP WiC SpP SuC AuC SpC WiP SpP SuC AuC WiC SpC WiP SpP SuC SpC AuC WiC SpP SuC AuC WSpC WiP WiC WD WD TGR TGR TGR EWW EWW EWW EWW EWW WS WS WS WS WS WS WD WD WD WD SST SST SST SST SST SST TGR TGR TGR EWW EWW WiP 1 0.3 0.03 0.08 0.1 0 0.72 0.14 0.17 0 0.11 -0.08 0.59 0.13 0.09 0.05 0.18 -0.12 0.05 -0.11 -0.11 -0.19 -0.07 -0.18 -0.29 -0.18 0.06 0.05 0.04 0 EWW SpP 1 -0.19 -0.09 -0.19 -0.29 0.17 0.66 -0.09 -0.23 0.05 -0.35 0.12 0.67 -0.1 -0.15 0.11 -0.35 -0.08 -0.25 -0.1 -0.09 -0.05 0.01 -0.17 -0.1 0.03 0.14 0.11 0.08 EWW SuC 1 0.27 0.04 0.09 0.02 -0.16 0.22 0.24 0.04 0.12 0.1 -0.15 0.16 0.39 0.13 0.16 0.07 0.08 -0.41 -0.33 -0.04 0.12 -0.02 -0.06 0.04 0.08 -0.07 0.05 EWW AuC 1 0.03 0.17 0.21 0.05 0.25 0.6 0.1 0.06 0.17 0.05 0.08 0.52 0.05 0 0.04 -0.03 -0.07 -0.48 -0.59 -0.16 0.14 -0.03 -0.14 -0.09 0.04 0.08 EWW WiC 1 0.28 -0.17 -0.35 -0.16 0.04 0.7 0.11 -0.03 -0.28 -0.11 0.1 0.57 0.13 0.28 0.11 0.01 0.15 0.12 -0.09 -0.05 -0.14 -0.11 -0.28 -0.34 -0.22 EWW WSpC 1 0.01 -0.15 0.06 -0.03 0.14 0.64 0.09 -0.08 0.08 0.13 0.09 0.65 0.13 0.21 -0.08 0.19 -0.05 -0.24 0 0 0.09 -0.19 -0.2 -0.12 WS WiP 1 0.41 0.5 0.26 -0.09 -0.06 0.74 0.19 0.3 0.14 -0.09 -0.21 -0.1 -0.06 0.05 -0.26 -0.13 -0.13 -0.04 0.05 0.12 0.18 0.18 0.14 WS SpP 1 0.31 0.01 -0.08 -0.1 0.16 0.84 0.09 -0.1 -0.15 -0.23 -0.01 -0.09 0.1 -0.1 -0.1 0.09 0.09 0.17 0.14 0.25 0.28 0.23 WS SuC 1 0.46 0.14 0.32 0.42 0.15 0.75 0.38 0.11 0.18 -0.04 0.14 -0.03 -0.18 -0.1 0.02 0.07 0.07 0.05 0.11 0.08 0.17 WS AuC 1 0.19 0.06 0.2 -0.11 0.2 0.74 0.07 -0.05 -0.04 -0.02 -0.13 -0.41 -0.37 0.09 0.12 0.11 -0.05 0.06 0.14 0.12 WS WiC 1 0.38 -0.01 -0.11 0 0.12 0.73 0.19 0.19 -0.1 -0.15 -0.09 -0.04 -0.03 0.04 0.07 0.07 0.01 -0.09 0.02 WS WSpC 1 -0.08 -0.11 0.19 0.03 0.16 0.85 0.09 0.06 -0.14 0.01 0.04 -0.07 0.21 0.28 0.29 0.15 0.07 0.15 WD WiP 1 0.21 0.54 0.47 0.34 -0.01 0.01 0.06 0.02 -0.14 -0.19 -0.09 -0.47 -0.41 -0.23 -0.04 -0.25 -0.33 WD SpP 1 0.13 0.07 0.05 -0.08 0.08 0.02 0.01 -0.08 -0.11 0.08 -0.17 -0.14 -0.02 0.06 -0.05 -0.11 WD SuC 1 0.42 0.29 0.25 -0.12 0.15 -0.04 -0.04 -0.13 -0.05 -0.22 -0.35 -0.27 -0.05 -0.2 -0.17

159 WD AuC 1 0.36 0.15 0.05 0.1 -0.15 -0.34 -0.37 0.01 -0.3 -0.34 -0.32 -0.21 -0.29 -0.28 WD WiC 1 0.23 0.16 -0.03 -0.12 0.01 0.04 0.07 -0.41 -0.42 -0.26 -0.22 -0.47 -0.42 WD SpC 1 0.11 0.12 -0.1 0.13 0.1 0.03 -0.02 0.02 0.1 0.02 -0.17 -0.15 SST WiP 1 0.56 0.28 0.31 0.05 0.14 -0.29 -0.14 -0.03 -0.18 -0.06 0.09 SST SpP 1 0.49 0.51 0.25 0.3 -0.09 -0.23 -0.19 -0.42 -0.16 -0.02 SST SuC 1 0.56 0.21 0.18 0.05 0 -0.27 -0.31 -0.03 0.01 SST AuC 1 0.49 0.19 -0.03 -0.11 -0.19 -0.45 -0.2 -0.1 SST WiC 1 0.55 0 0.07 0.12 -0.12 -0.26 -0.11 SST SpC 1 -0.09 -0.08 -0.11 0.02 -0.07 -0.22 TGR WiP 1 0.77 0.46 0.29 0.46 0.55 TGR SpP 1 0.81 0.62 0.57 0.6 TGR SuC 1 0.7 0.47 0.53 TGR AuC 1 0.57 0.37 TGR WiC 1 0.76 TGR SpC 1 Table C.6: p-values for correlation coefficients for wind and sea temperature covariates. WiP SpP SuC AuC WiC SpC SpC AuC WiC WiP SpP SuC WiP SpP SuC AuC WiC SpC AuC SpC WiP SpP SuC WiC WiC WiP SpP SuC SpC AuC TGR WD WD WD WD WD TGR TGR TGR EWW EWW EWW EWW EWW EWW WS WS WS WD SST SST SST SST SST SST TGR TGR WS WS WS EWW WiP 1.00 0.05 0.82 0.61 0.52 0.98 0.00 0.37 0.25 0.99 0.45 0.58 0.00 0.38 0.57 0.73 0.24 0.41 0.73 0.47 0.47 0.20 0.64 0.24 0.05 0.23 0.69 0.77 0.80 0.99 EWW SpP 1.00 0.21 0.54 0.22 0.06 0.27 0.00 0.56 0.12 0.75 0.02 0.44 0.00 0.52 0.34 0.48 0.02 0.58 0.10 0.50 0.54 0.75 0.95 0.28 0.52 0.87 0.36 0.46 0.61 EWW SuC 1.00 0.08 0.78 0.55 0.91 0.30 0.15 0.12 0.80 0.45 0.52 0.34 0.31 0.01 0.41 0.28 0.64 0.59 0.01 0.03 0.77 0.43 0.87 0.70 0.81 0.60 0.63 0.74 EWW AuC 1.00 0.84 0.27 0.16 0.74 0.10 0.00 0.52 0.72 0.26 0.76 0.61 0.00 0.76 1.00 0.81 0.86 0.63 0.00 0.00 0.31 0.37 0.85 0.37 0.55 0.78 0.58 EWW WiC 1.00 0.07 0.27 0.02 0.30 0.80 0.00 0.48 0.86 0.06 0.47 0.52 0.00 0.41 0.07 0.47 0.95 0.32 0.45 0.57 0.74 0.37 0.48 0.06 0.02 0.15 EWW WSpC 1.00 0.93 0.32 0.70 0.84 0.37 0.00 0.56 0.58 0.59 0.39 0.55 0.00 0.38 0.17 0.59 0.22 0.73 0.12 0.99 1.00 0.54 0.21 0.20 0.45 WS WiP 1.00 0.01 0.00 0.08 0.58 0.69 0.00 0.20 0.05 0.34 0.57 0.17 0.49 0.71 0.77 0.09 0.38 0.38 0.78 0.74 0.41 0.23 0.24 0.36 WS SpP 1.00 0.04 0.97 0.60 0.53 0.29 0.00 0.54 0.51 0.34 0.13 0.96 0.55 0.52 0.52 0.53 0.57 0.54 0.27 0.37 0.10 0.06 0.14 WS SuC 1.00 0.00 0.37 0.03 0.00 0.34 0.00 0.01 0.46 0.23 0.81 0.36 0.86 0.24 0.52 0.88 0.66 0.63 0.73 0.46 0.59 0.28 WS AuC 1.00 0.21 0.67 0.19 0.48 0.18 0.00 0.65 0.74 0.78 0.87 0.41 0.01 0.01 0.55 0.43 0.46 0.76 0.70 0.37 0.45 WS WiC 1.00 0.01 0.94 0.47 0.98 0.45 0.00 0.22 0.21 0.50 0.34 0.56 0.82 0.87 0.80 0.64 0.66 0.96 0.54 0.91 WS WSpC 1.00 0.62 0.48 0.22 0.87 0.28 0.00 0.54 0.71 0.36 0.96 0.81 0.63 0.17 0.06 0.06 0.31 0.64 0.33 WD WiP 1.00 0.16 0.00 0.00 0.02 0.93 0.97 0.71 0.90 0.37 0.22 0.54 0.00 0.01 0.13 0.78 0.10 0.02 WD SpP 1.00 0.41 0.66 0.75 0.60 0.60 0.88 0.93 0.59 0.47 0.58 0.28 0.35 0.88 0.67 0.75 0.48 WD SuC 1.00 0.00 0.05 0.10 0.44 0.31 0.80 0.78 0.38 0.74 0.14 0.02 0.07 0.75 0.19 0.27 WD AuC 1.00 0.01 0.33 0.77 0.52 0.34 0.02 0.01 0.95 0.04 0.02 0.03 0.18 0.05 0.06 WD WiC 1.00 0.12 0.30 0.86 0.43 0.97 0.81 0.65 0.01 0.00 0.08 0.15 0.00 0.00 160 WD SpC 1.00 0.47 0.45 0.49 0.38 0.52 0.84 0.88 0.90 0.51 0.92 0.27 0.32 SST WiP 1.00 0.00 0.06 0.04 0.77 0.36 0.05 0.37 0.87 0.23 0.70 0.57 SST SpP 1.00 0.00 0.00 0.10 0.05 0.55 0.13 0.20 0.00 0.28 0.88 SST SuC 1.00 0.00 0.17 0.24 0.73 0.98 0.07 0.04 0.85 0.95 SST AuC 1.00 0.00 0.21 0.83 0.49 0.20 0.00 0.20 0.50 SST WiC 1.00 0.00 0.98 0.67 0.43 0.41 0.09 0.47 SST SpC 1.00 0.54 0.62 0.49 0.92 0.63 0.15 TGR WiP 1.00 0.00 0.00 0.06 0.00 0.00 TGR SpP 1.00 0.00 0.00 0.00 0.00 TGR SuC 1.00 0.00 0.00 0.00 TGR AuC 1.00 0.00 0.01 TGR WiC 1.00 0.00 TGR SpC 1.00 C.3 Chlorophyll-a concentration

Table C.7: Chlorophyll-a covariates.

Covariate label Description Box1 Mean chlorophyll-a concentration in Box 1 (waters immedi- ately surrounding Phillip Island) Box2 Mean chlorophyll-a concentration in Box 2 (the region of the Bonney Upwelling) SuC Current summer AuC Current autumn WiC Current winter SpC Current spring

161 1.5 3.5 1.0 2.5 10 20 12 15

● ● ● ● ● ● ● ● ● ● ● ● ● ●

Box1_SuC ● ● ● ● ● ● ● ●●● ●● ●● ● ● ● ● ● ● ● ●● ●● ● ●●● ● ●●● ● ● ● ● ● ●● ● ● ●●●●● ● ●●● ●● ● ●● ● ●● ●● ● ● ● ● ●● ● ● 1.0 2.5

● ● ● ● ● ● ●

● ● ● ● ● ● ● Box1_AuC ●● ●● ● ● ● ● ●● ●● ● ● ●●● ● ●●● ●●● ● ●●●● ● ● ●● ● ● ●●● ● ● ●●● ●●● ●●● ● ●● ● ● ● ● ● ● ●● ● ●●●● ●● ● 1.5 3.5 ● ● ● ● ● ● ●

Box1_WiC ● ● ● ●● ● ●● ● ● ●● ● ● ● ● ●● ● ● ● ●●● ●●●● ●●●● ● ●● ● ● ●● ● ●●● ● ● ●●● ● ●● ●● 2 4 ●● ●● ●● ● ● ● ● ● ● ● ●● ●● ●

● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ●●● ●●● Box1_SpC ●● ● ● ●●● ● ● ●● ● ●● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ●● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 1.0 2.5 ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ●● ●● ● ● ● ● ● ● ●● ●● Box2_SuC ● ● ● ● ● ● ● ● ● ● ● ● ● ●●● ●●● ● ●● ● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 10 18 ● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ● ●● ●●● ● ● ● ● ● ●●● ●●● ●●● ●● ● ●● ●● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● Box2_AuC ●● ● ● ●● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 10 20 ● ● ● ● ● ● ●

●● ● ●●● ●●● ●● ● ● ● ● ● ● ●●● Box2_WiC ●● ● ● ●●● ●● ●● ●● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● ● ● ● ● ● ● ● ● ● ●● ●● ● ● ● ● ● ●

● ● ● ● ● ● ● 12 18

●● ●● ●● ●● ● ●● ● ● ● ● ●●● ●● ●● ●● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ●● ● ● ●● ●● ● Box2_SpC ● ● ● ● ● ● ● ●● ● ●●● ●● ● ●● ● ● ●● ●●● ● ● ●

12 15 ● ● ● ● ● ● ●

1.0 2.5 2 4 10 18 12 18

Figure C.4: Scatterplot matrix for chlorophyll-a covariates.

162 Box1_SuC Box1_AuC Box1_WiC Box1_SpC Box2_SuC Box2_AuC Box2_WiC Box2_SpC 1 Box1_SuC 0.8

Box1_AuC 0.6

Box1_WiC 0.4

0.2 Box1_SpC

0 Box2_SuC −0.2

Box2_AuC −0.4

Box2_WiC −0.6

−0.8 Box2_SpC

−1

Figure C.5: Correlation matrix for chlorophyll a covariates. The areas of the circles show the absolute value of the corresponding correlation coefficients.

163 Table C.8: Correlation coefficients for chlorophyll-a covariates. SuC AuC WiC SpC SuC AuC WiC SpC x1 x2 x1 x1 x1 x2 x2 x2 Bo Bo Bo Bo Bo Bo Bo Bo Box1 SuC 1.00 0.95 0.84 0.37 -0.18 -0.38 0.49 -0.32 Box1 AuC 1.00 0.90 0.47 -0.21 -0.25 0.64 -0.23 Box1 WiC 1.00 0.68 -0.23 -0.08 0.85 0.09 Box1 SpC 1.00 -0.26 0.18 0.70 0.50 Box2 SuC 1.00 0.48 -0.22 -0.10 Box2 AuC 1.00 0.20 0.38 Box2 WiC 1.00 0.29 Box2 SpC 1.00

Table C.9: p-values for correlation coefficients for chlorophyll-a covariates. SuC AuC WiC SpC SuC AuC WiC SpC x1 x2 x1 x1 x1 x2 x2 x2 Bo Bo Bo Bo Bo Bo Bo Bo Box1 SuC 1.00 0.00 0.00 0.22 0.56 0.20 0.09 0.29 Box1 AuC 1.00 0.00 0.11 0.50 0.41 0.02 0.44 Box1 WiC 1.00 0.01 0.46 0.80 0.00 0.77 Box1 SpC 1.00 0.40 0.57 0.01 0.08 Box2 SuC 1.00 0.10 0.48 0.75 Box2 AuC 1.00 0.52 0.21 Box2 WiC 1.00 0.34 Box2 SpC 1.00

164 Appendix D

Principal Component Analysis

Principal Component Analysis (PCA) is an approach that can be used to reduce a large set of variables, where more than one variable might be measuring the same driving principle governing the behaviour of a system, to a smaller set which still contains most of the information of the large set. PCA generates a new set of ‘synthetic’ variables called principal components (PCs), each of which is a linear combination of the original variables. The PCs account for the greatest amount of the variability in the data.

The first principal component is the direction through the data that explains the most variability in the data. The second and subsequent PCs are computed un- der the constraint of being orthogonal to the first so that there is no redundant information. Each subsequent PC describes the greatest amount of the remaining variability. [Abdi and Williams, 2010].

D.1 Advantages and disadvantages

With a large dataset, a natural instinct is to try to reduce the number of variables to a number that will still convey virtually all the information in the original variables, and PCA is one of the simplest ways to achieve this. Although this approach is less intuitive than choosing a subset of variables, it has the advantage that a greater reduction in dimensionality can be achieved for the same amount of information loss [Jolliffe, 1990]. An advantage of PCA is that it allows combining variables of different nature, for example, temperature and rainfall.

Generally, the PCs with higher eigenvalues are selected as those contributing most to the total variation. However, variables that could be of biological importance may not be accounted for by these PCs, resulting in the loss of essential information. Thus it is important to select PCs in such a way that all original covariates are well represented, since a ‘good’ PC may not be a true reflection of the system under

165 consideration and discarded PCs may be important contributors to the fit of the response variable [Grosbois et al., 2008, Hadi and Ling, 1998, Jolliffe, 1982].

Each PC is a linear combination of the original variables, which can make them difficult to understand and interpret. Additionally, since only linear relationships are considered, this method does not take into account potential higher-order inter- actions between variables [Jolliffe, 1993]. In many applications it may be desirable not only to reduce the number of dimensions but also the number of variables. Mc- Cabe [1984] presents an alternative methodology whereby principal variables are selected, making them easier to deal with than linear combinations of variables.

D.2 Method and analysis

To illustrate the method, I carried out a PCA in R (version 3.3.1, 21-06-2016), using the 10 variables from the air temperature and rainfall analysis (Table D.1).

Table D.1: Variables used in PCA.

Label Description RB Total rainfall in breeding season RM Total rainfall in moult period TB Mmean ambient air temperature in breeding season TM Mean ambient air temperature in moult period D27B Number of degree days over 27◦C in breeding season D27M Number of degree days over 27◦C in moult period D35B Number of degree days over 35◦C in breeding season D35M Number of degree days over 35◦C in moult period VB Mean vapour pressure in breeding season VM Mean vapour pressure in moult period

166 Output from the analysis is shown below.

#Standard deviations: # [1] 1.7938002 1.4846883 1.1672910 1.1086927 0.9776694 0.5845545 0.5091616 # [8] 0.4482679 0.3893478 0.2772921

#Rotation: (loadings are [proportional to the] coefficients in the linear combination of original variables that make up the PCs). Loadings are analogous to correlation coefficients; squaring them gives the amount of explained variation.

# PC1 PC2 PC3 #RB 0.26377378 -0.413997431 0.07575167 #RM 0.07564624 -0.255030222 0.12795183 #TB -0.50682642 -0.131556720 0.10216995 #TM -0.39965388 -0.083552943 -0.43733271 #D27B -0.47318852 -0.088423040 0.34021029 #D27M -0.24785518 0.454171569 -0.34053250 #D35B -0.36978157 -0.007083158 0.54782014 #D35M -0.13470789 0.382591879 -0.02097095 #VB -0.13060453 -0.489222919 -0.11984818 #VM -0.22543971 -0.373391508 -0.47914061

# PC4 PC5 PC6 #RB -0.41057735 0.301725207 -0.06101028 #RM -0.13772921 -0.897833863 0.16895315 #TB -0.08273287 -0.062560737 0.32898820 #TM 0.22960420 0.104932896 0.35380760 #D27B -0.05970457 0.009167587 -0.21535678 #D27M -0.27937952 -0.088091840 0.09384996 #D35B 0.13332059 0.103135580 -0.18811721 #D35M -0.65677218 -0.065520955 -0.28409669 #VB -0.45518794 0.198204246 0.25881134 #VM 0.12229630 -0.160849034 -0.70430877

# PC7 PC8 PC9 PC10 #RB 0.42984876 0.23211559 0.47632939 -0.1562683 #RM 0.18873237 -0.07570001 0.08751442 0.1011743 #TB -0.13855189 0.41607547 -0.12186936 -0.6238764 #TM 0.55280301 0.10289042 -0.03307461 0.3685857 #D27B -0.31399555 0.28116177 0.36887117 0.5373943 #D27M -0.07938699 -0.35400279 0.59512346 -0.1895270 #D35B 0.42532348 -0.53380983 -0.04487676 -0.1773081 #D35M 0.26349004 0.19503038 -0.44168832 0.1270095 #VB -0.31426669 -0.47745626 -0.24773144 0.1663629 #VM -0.02891742 -0.03691118 -0.02818196 -0.2007672

167 More detail can be obtained using the summary(pca.data2) command.

# PC1 PC2 PC3 PC4 PC5 #Standard deviation 1.7938 1.4847 1.1673 1.1087 0.97767 #Proportion of Variance 0.3218 0.2204 0.1363 0.1229 0.09558 #Cumulative Proportion 0.3218 0.5422 0.6785 0.8014 0.89696

# PC6 PC7 PC8 PC9 PC10 #Standard deviation 0.58455 0.50916 0.44827 0.38935 0.27729 #Proportion of Variance 0.03417 0.02592 0.02009 0.01516 0.00769 #Cumulative Proportion 0.93113 0.95706 0.97715 0.99231 1.00000

The linear combination for the first PC, using the loadings for each of the variables, is

# RB RM TB TM D27B # 0.26377378 0.07564624 -0.50682642 -0.39965388 -0.47318852 D27M D35B D35M VB VM -0.24785518 -0.36978157 -0.13470789 -0.13060453 -0.22543971

For this PC, we note that the coefficients for the rainfall variables are positive and those of the temperature variables are negative. This indicates that this PC best describes conditions of high rainfall and low temperature, that is, wet and cold. The linear combination for the second PC is

# RB RM TB TM D27B #-0.413997431 -0.255030222 -0.131556720 -0.083552943 -0.088423040 # D27M D35B D35M VB VM #0.454171569 -0.007083158 0.382591879 -0.489222919 -0.373391508

For this PC, the coefficients of the rainfall variables are negative, and those of the temperature variables are a mixture of positive and negative. It is more difficult to biologically interpret this PC — it represents conditions of low rainfall, but is less useful in describing temperature. The coefficients for prolonged high temperatures during the moult period are positive, so this PC might work best in conditions of low rainfall and high prolonged temperatures during moult. The linear combination for the third PC is

# RB RM TB TM D27B # 0.07575167 0.12795183 0.10216995 -0.43733271 0.34021029 # D27M D35B D35M VB VM # -0.34053250 0.54782014 -0.02097095 -0.11984818 -0.47914061

This PC has positive coefficients for rainfall, with positive coefficients for temper- ature during the breeding season and negative coefficients for rainfall during the moult period. Again, it is difficult to interpret this PC — it represents wet and

168 hot conditions during the breeding season, and cooler conditions during the moult period.

Figure D.1 shows the variances accounted for by each of the PCs. The first two PCs account for 54.2% of the total variance and the first five PCs account for 89.7% of the variance.

Figure D.1: Variances of the principal components for air temperature and rainfall variables.

D.3 Using the PCs in survival models

To create the covariates for use in the survival models, the original data were re- scaled according to the linear combination of the PCs. For example, PC1 = ax1 + bx2 +...+jx10 where the a . . . j are the coefficients for each of the original variables x1 . . . x10. It is helpful to think of the coefficients as ‘weightings’ for each of the original variables based on their contribution to the total variance.

169 D.3.1 Terrestrial covariates (Chapter 5)

Here I used the PCs as calculated above for the air temperature and rainfall covari- ates.

1. Using PC1 as a covariate at the first stage yielded an AIC of 633.86 which is significantly higher than the best first-stage model when using the original covariates (hum mlt, AIC = 553.27) indicating no improvement in the model fit. 2. Using PC2 yielded an AIC of 595.80 which again is significantly higher than the best first-stage model, indicating no improvement in the model fit. 3. Using PC3 yielded an AIC of 585.72, a relatively small change from when PC2 was used.

Note that the AIC for the survival model using PC2 was lower than that obtained when using PC1. This is because PCA is an unsupervised algorithm method and hence the PCs with high variances could simply indicate noise rather than a signal.

D.3.2 Wind and sea temperature (Chapter 6)

PCA was carried out using all 30 wind and sea temperature variables. Given the high correlations between many of the variables, I did not discard any before the PCA. From Figure D.2, I considered the first three PCs. The linear combination for the first PC, using the loadings for each of the variables, is

# EWW_WiP EWW_SpP EWW_SuC EWW_AuC EWW_WiC EWW_WSpC #-0.064220994 0.066165888 -0.066510824 -0.054995668 -0.181870055 -0.109417271 # WS_WiP WS_SpP WS_SuC WS_AuC WS_WiC WS_WSpC # 0.010012321 0.128437100 -0.052774286 -0.042533412 -0.094225914 0.003143000 # WD_WiP WD_SpP WD_SuC WD_AuC WD_WiC WD_SpC #-0.223180790 -0.005816244 -0.192554352 -0.246798950 -0.288526954 -0.103277163 # SST_WiP SST_SpP SST_SuC SST_AuC SST_WiC SST_SpC #-0.111769287 -0.152847901 -0.041680928 -0.084934144 -0.010110483 -0.049008582 # TGR_WiP TGR_SpP TGR_SuC TGR_AuC TGR_WiC TGR_SpC # 0.299320355 0.365112919 0.313837017 0.291397245 0.337467151 0.315716799

This PC illustrates one of the shortcomings of this method which is the difficulty in interpreting it in a biologically meaningful way.

Using the PCs as covariates in models gives the following results for first-year and adult survival models. Note the best first-year covariate using the original variables was EWW SpP (AIC = 414.97).

170 Figure D.2: Variances of the principal components for 30 wind and sea temperature variables.

First-year survival

1. Using PC1 as a covariate at the first stage yielded an AIC of 457.63 which was significantly higher than the best first-stage model when using the original covariates. 2. Using PC2 yielded an AIC of 480.98 which again was significantly higher than the best first-stage model. 3. Using PC3 yielded an AIC of 558.24, again, significantly higher that the best first-stage model.

Adult survival

The best adult covariate was the residuals of the sea temperature gradient in the previous winter (AIC = 548.24).

1. Using PC1 as a covariate at the first stage yielded an AIC of 633.86 which was significantly higher than the best first-stage model when using the original covariates.

171 2. Using PC2 yielded an AIC of 595.80 which was an improvement on the first- stage model using PC1, but not as good as the first-stage model using the original covariates. 3. Using PC3 yielded an AIC of 585.72, which, although better than when either PC1 or PC2 were used, was still significantly higher that the best first-stage model using the original covariates.

D.3.3 Chlorophyll-a concentration (Chapter 7)

First-year survival

PCA was carried out using all eight chlorophyll a variables.

From Figure D.3, I considered the first three PCs.

The linear combination for the first PC, using the loadings for each of the variables, is

Box1_SuC Box1_AuC Box1_WiC Box1_SpC 0.43337840 0.46650649 0.49631470 0.36340495 Box2_SuC Box2_AuC Box2_WiC Box2_SpC -0.17678391 -0.08513377 0.42056631 0.02805506 The linear combination for the second PC, using the loadings for each of the vari- ables, is

Box1_SuC Box1_AuC Box1_WiC Box1_SpC -0.29178715 -0.19346676 0.04235613 0.37174001 Box2_SuC Box2_AuC Box2_WiC Box2_SpC 0.13112912 0.54891643 0.27110405 0.58774544

Using PC1 at the first-stage yielded an AIC of 175.27, PC2 yielded an AIC of 210.11 and PC3 yielded an AIC of 220.78. Note that the AIC for the model using PC1 is better than that using the original covariate (Box1 SuC, 188.26) but not as good as the quadratic model (Box2 SuC quad, 153.29).

172 Figure D.3: Variances of the principal components for chlorophyll-a variables used in first-year survival models.

Adult survival

PCA was carried out using the four variables used in adult survival models.

From Figure D.4, I considered the first PC.

The linear combination for the first PC, using the loadings for each of the variables, is

Box1_SuC Box1_AuC Box1_WiC Box1_SpC 0.5162011 0.5396616 0.5470812 0.3781586

Using PC1 at the first stage yielded an AIC of 58.66 which was higher than the AIC of the best first-stage model using the original variables (Box1 SuC, 49.39).

173 Figure D.4: Variances of the principal components for chlorophyll-a variables used in adult survival models.

In this appendix, I carried out a PCA on all the covariates used in my thesis, and have shown some of the shortcomings of this approach. When used in survival models, none of the PCs gave better fits to the models than the original covariates. I concluded that the methods used in my thesis gave better fits of the models to the data and allowed for more useful interpretations of the results.

174 Appendix E

LOC and SITE Codes used in this study

Phillip Island location (LOC) and site (SITE) codes.

LOC SITE Description 1 (Parade) 15 1 31 (night), 32 (day) Roundup (PSG), discontinued 30 unspecified 16 Hospital 18 Car Park 3 (Cliffs to West) 14 3 43 RT Bay 26 General 45 Nobbies site 4 (North Shore) 21 22 Wiped out by foxes 23 33 24 25 27 Cowries/General 28 (Summerland East North) 28 Nobbies side to Mandeville Rd 29 Parade side to Ventnor Rd 29 (Summerland East South) 28 Nobbies side 29 To St Leonard’s

175

Appendix F

COM, METH and STATUS Codes

F.1 COM codes

008 Extra food (control) 013 Band only in transponder experiment 015 Band wear 101 Yes (alert that there is an entry in the field data book) 102 ExA stomach flushing 103 ExR radio-tracking 104 Translocation 105 ExB CuSulphate 106 ExW Injected and bled for metabolism 107 ExA + ExR 108 Extra food (given food) 109 Drenched 110 Westernport dump 111 Experiment – other institution 112 Rebanded 113 Tag only in transponder experiment 114 Fix band 115 Band wear 116 Bled for genetic study 117 30 burrows human disturbance experiment 118 Additional 20 burrows human disturbance experiment 119 Depth gauge experiment 120 Nobbies relocation 121 Student project 122 Returned alive from scientific study 123 Rabbit Island darkened bands (2001) 124 Stomach flushed plus blood taken

177 125 Carpark relocation project 126 Band recovery information unknown 127 Shelly Beach tour chicks disturbed 128 Heart rate logger inserted 130 Cat Bay/Shellhouse study (Rebecca) 201 In rehabilitation then released 203 Satellite tracking 204 Translocation recovery 208 Food given recovery 209 Drenched bird recovery 210 Western Port dump recovery 213 Tag plus band in transponder experiment 220 Accidentally transpondered twice 221 Area 3 Andre transponders 222 Died during scientific study 228 Heart rate logger removed 303 GPS tracking 308 Control recovery 413 Disqualified

178 F.2 Methods of Encounter Codes (METH)

00 METHOD NOT COVERED BY CODE 01 PROBABLY TRAPPED DEVICE UNKNOWN 02 TRAPPED DEVICE UNKNOWN 03 TRAPPED IN MIST NET 04 TRAPPED IN CAGE TRAP 05 TRAPPED WITH CANNON NET 06 TRAPPED IN CLAP TRAP, SPRUNG TRAP, ETC 07 TRAPPED WITH BAL-CHATRI/NOOSE CARPET 08 TRAPPED BY HAND OR WITH HANDHELD NET 09 TRAPPED USING LIGHT DEVICE 0A TRAPPED WITH A DHO-GHAZA 0R LOCATED USING RADIO TELEMETRY 10 TRAPPED IN HARP TRAP 11 TRAPPED IN MONOFILAMENT MIST NETS 12 TRAPPED WITH TRIP WIRE OVER WATER 1A HAND CAUGHT AT NEST 13 HAND CAUGHT AT ROOST OR NEST 14 TRAPPED AS ATTRACTED TO DOMESTIC BIRDS 15 DELIBERATELY TRAPPED FOR THE AVIARY 16 TRAPPED BECAUSE BAND TANGLED IN NATURAL OBJECT 17 TRAPPED BECAUSE BAND TANGLED IN HUMAN OBJECT 18 TRAPPED BECAUSE BAND TANGLED IN FISHING GEAR 19 TRAPPED BECAUSE BIRD TANGLED IN NATURAL OBJECT 20 TRAPPED BECAUSE BIRD TANGLED IN HUMAN OBJECT 21 TRAPPED BECAUSE BIRD TANGLED IN FISHING GEAR 2A CAUGHT ON LONGLINE 22 TRAPPED ACCIDENTALLY IN TRAP FOR TERRESTRIAL ANI- MALS 23 TRAPPED ACCIDENTALLY IN MARINE/AQUATIC ANIMAL TRAP 24 TRAPPED USING NARCOTIC DRUGS 25 FOUND SICK OR INJURED 26 EXHAUSTED 27 INJURED BY BAND 28 OILED 29 BURNT OR SCORCHED BY FIRE 30 FOUND NEAR ELECTRICITY WIRES 31 COLLIDED WITH A MOVING ROAD VEHICLE

179 32 COLLIDED WITH A MOVING TRAIN 33 COLLIDED WITH A MOVING AIRCRAFT 34 COLLIDED WITH A MOVING SHIP 35 COLLIDED WITH A LIGHTHOUSE OR STATIONARY NIGHT LIGHT 36 COLLIDED WITH A WINDOW OR OTHER TRANSPARENT MA- TERIAL 37 COLLIDED WITH A BUILDING, NON-WIRE FENCE, IMMOBILE VEHICLE 38 COLLIDED WITH A MAST, TOWER, POLE, WIRE FENCE, AERIAL SPRINKLER 39 FOUND ON HIGHWAY/ROAD; BUT NOT CERTAINLY HIT BY CAR 3A COLLIDED WITH A NATURAL OBJECT EG TREE, CLIFF 40 BAND FOUND ON BIRD, NO FURTHER DATA ON METHOD OF ENCOUNTER 41 BAND RETURNED, NOT REPORTED IF BAND ON BIRD 42 BAND ONLY FOUND (USE STATUS=02) 43 BAND NUMBER ONLY REPORTED 44 BAND LOST 45 BAND DESTROYED OR DAMAGED 46 COLOUR MARKING SIGHTED IN FIELD (COHORT ONLY) (USE STATUS=26) 47 BAND NUMBER READ IN FIELD (BIRD NOT TRAPPED) (USE STATUS=26) 48 COLOUR MARKING SIGHTED IN FIELD (BAND NO. INFERRED) (USE STATUS=26) 49 BAND NUMBER/COLOUR MARKING SIGHTED ON BIRD ON NEST (USE STATUS=26) 50 CAPTIVE BRED BIRD/BAT 51 SUCKLING YOUNG HAND RAISED (BATS ONLY) 52 NESTLING HAND RAISED (ABANDONED, ORPHANED OR NEST DESTROYED) 53 BANDED AFTER DEATH FOR EXPERIMENT 54 FOUND FLOATING IN SEA OR FRESHWATER OR BEACH- WASHED 55 FOUND IN/ON CAR, SHIP, ETC PROBABLY ENCOUNTERED ELSEWHERE 56 TRAPPED/KILLED BECAUSE IT WAS BANDED

180 57 BAND FOUND ON SPECIES DIFFERENT TO THAT BANDED (SEE FILE) 58 LEG (OR WING) AND BAND ONLY FOUND (USE STATUS=02) 59 COLOUR MARKER FOUND NOT ON BIRD (BAND NO. IN- FERRED) 5A COLOUR MARKER FOUND ON WING ONLY (BAND NO. IN- FERRED) (USE STATUS=05) 60 READABLE BAND SIGHTED, NO. ON STANDARD BAND IN- FERRED (USE STATUS=26) 61 SHOT (REASON UNKNOWN) 62 SEIZED FOR LAW ENFORCEMENT REASONS 63 TAKEN FOR SCIENTIFIC STUDY (NOT BANDING) 64 TAKEN TO PROTECT CROPS 65 TAKEN TO PROTECT DOMESTIC ANIMALS 66 TAKEN FOR AIRCRAFT STRIKE PREVENTION PROGRAM 67 TAKEN FOR FOOD, FEATHERS, CEREMONIAL REASONS 68 SHOT FOR SPORT/FOOD 69 TAKEN FOR HUMAN HEALTH REASONS 6A TAKEN FOR NATURE CONSERVATION REASONS 70 SHOT WITH ARROW OR SPEARED 71 ACCIDENTLY INJURED/KILLED IN EXPLOSION 72 POISONED - UNKNOWN IF INTENTIONAL 73 POISONED NATURAL SOURCE 74 UNINTENTIONALLY POISONED BY BAIT FOR OTHER ANIMALS 75 UNINTENTIONALLY POISONED BY AERIAL SPRAYING OF CROPS 76 UNINTENTIONALLY POISONED BY INDUSTRIAL WASTES 77 INTENTIONALLY POISONED WITH BAITS 78 INTENTIONALLY POISONED BY AERIAL SPRAYING FOR BIRDS 79 LEAD POISONED (LEAD SHOT) 80 TAKEN BY UNKNOWN ANIMAL 81 TAKEN BY DOMESTIC OR WILD CAT 82 TAKEN BY DOMESTIC OR WILD DOG 83 TAKEN BY DOMESTIC ANIMAL (SPECIES?) 84 TAKEN BY A WILD MAMMAL (SPECIES?) 85 TAKEN BY A WILD BIRD (SPECIES?) 86 TAKEN BY A WILD FISH (SPECIES?)

181 87 TAKEN BY A WILD REPTILE (SPECIES?) 88 CARCASS BEING EATEN BY SCAVENGING BIRDS 89 FOUND DEAD OR INJURED AFTER A STORM 90 OBSOLETE CODE 91 OBSOLETE CODE 92 INJURED OR KILLED BY HUMAN (NOT FOR FOOD) 93 INJURED/DIED DURING EXPERIMENTAL ACTIVITIES 94 ELECTROCUTED 95 FOUND IN STILL WATER 96 CAPTIVE BIRD/BAT (WAS FROM WILD) (USE STATUS=13) 97 FOUND INSIDE A MAN MADE STRUCTURE 98 FOUND DEAD IN/NEAR A NEST (PULLI AND ADULTS) 99 FOUND DEAD, CAUSE UNKNOWN 9A BANDING DATA UNKNOWN (SEE ‘NO SCHEDULES’ FILE)

182 F.3 Status codes

00 STATUS OF BIRD/BAT AND BAND IS UNKNOWN 01 STATUS OF BIRD/BAT IS UNKNOWN AND THE BAND WAS LEFT ON 02 STATUS OF BIRD/BAT IS UNKNOWN AND THE BAND WAS RE- MOVED 03 WAS DEAD AND THE STATUS OF THE BAND IS UNKNOWN 04 WAS DEAD AND THE BAND WAS LEFT ON 05 WAS DEAD AND THE BAND WAS REMOVED 06 WAS MERCY KILLED AND THE STATUS OF THE BAND IS UN- KNOWN 07 WAS MERCY KILLED AND THE BAND WAS LEFT ON 08 WAS MERCY KILLED AND THE BAND WAS REMOVED 09 REHABILITATION ATTEMPTED BUT BIRD/BAT DIED, STATUS OF BAND UNKNOWN. 10 REHABILITATION ATTEMPTED BUT BIRD/BAT DIED, BAND LEFT ON 11 REHABILITATION WAS ATTEMPTED BUT BIRD/BAT DIED, BAND WAS REMOVED 12 WAS RELEASED ALIVE, STATUS OF BAND IS UNKNOWN 13 WAS RELEASED ALIVE WITH THE BAND 14 WAS RELEASED ALIVE AND THE BAND WAS REPLACED DUE TO WEAR, ETC 15 WAS REHABILITATED & RELEASED ALIVE, BAND STATUS IS UNKNOWN 16 WAS REHABILITATED & RELEASED ALIVE WITH THE BAND 17 WAS REHABILITATED & RELEASED ALIVE, BAND WAS RE- MOVED 18 IS ALIVE IN CAPTIVITY AND STATUS OF BAND IS UNKNOWN 19 IS ALIVE IN CAPTIVITY WITH BAND 20 IS ALIVE IN CAPTIVITY AND BAND WAS REMOVED 21 ALIVE: UNKNOWN IF RELEASED OR CAPTIVE, BAND STATUS UNKNOWN 22 ALIVE BUT UNKNOWN IF RELEASED OR CAPTIVE, BAND WITH BIRD/BAT 23 ALIVE: UNKNOWN IF RELEASED OR CAPTIVE, BAND RE- MOVED 24 TRANSPORTED TO NEW SITE AND RELEASED WITH BAND 25 TRANSPORTED TO NEW SITE AND BAND REMOVED

183 26 WAS ALIVE IN THE WILD WITH THE BAND 27 PARTIALLY DECOMPOSED AND BAND STATUS UNKNOWN 28 PARTIALLY DECOMPOSED AND BAND LEFT ON 29 PARTIALLY DECOMPOSED AND BAND REMOVED 30 WAS SKELETON/DRIED OUT CORPSE, BAND STATUS UN- KNOWN 31 WAS SKELETON/DRIED OUT CORPSE, BAND LEFT ON 32 WAS SKELETON/DRIED OUT CORPSE, BAND REMOVED 33 FLEW AWAY WITHOUT THE BAND 34 RELEASED ALIVE & THE BAND WAS REPLACED DUE TO IN- JURY 35 STATUS OF BIRD/BAT UNKNOWN 36 RELEASED ALIVE, BAND REMOVED AND NOT REPLACED DUE TO INJURY 37 RELEASED ALIVE CARRYING 2 OR MORE BANDS 38 DIED DURING CAPTURE, BAND STATUS IS UNKNOWN 39 DIED DURING CAPTURE, BAND LEFT ON 40 DIED DURING CAPTURE, BAND REMOVED 41 KILLED IN NET BY PREDATOR, BAND STATUS IS UNKNOWN 42 KILLED IN NET BY PREDATOR, BAND LEFT ON 43 KILLED IN NET BY PREDATOR, BAND REMOVED 44 RELEASED ALIVE AND THE BAND WAS REMOVED 45 RELEASED ALIVE WITH BAND AND ELECTRONIC TAG 47 RELEASED ALIVE AND BAND REMOVED & REPLACED WITH A MICROCHIP 99 DIED BEFORE BANDING

184 References

H. Abdi and L. J. Williams. Principal component analysis. Wiley Interdisciplinary Reviews: Computational Statistics, 2(4):433–459, 2010. J. Adams and S. Flora. Correlating seabird movements with ocean winds: linking satellite telemetry with ocean scatterometry. Marine Biology, 157(4):915–929, 2010. I. Af´an,A. Chiaradia, M. G. Forero, P. Dann, and F. Ram´ırez. A novel spatio- temporal scale based on ocean currents unravels environmental drivers of repro- ductive timing in a marine predator. Proceedings of the Royal Society of London B: Biological Sciences, 282(1810):20150721, 2015. P. Agnew, D. Houston, C. Lalas, and J. Wright. Variation in reproductive per- formance of little penguins (Eudyptula minor) attributable to double brooding. Journal of Ornithology, 155(1):101–109, 2014. P. Agnew, C. Lalas, J. Wright, and S. Dawson. Annual variation in recruitment and age-specific survival of little penguins, Eudyptula minor. Emu, 116(1):62–70, 2016. D. G. Ainley, W. J. Sydeman, and J. Norton. Upper trophic level predators indicate interannual negative and positive anomalies in the California Current food web. Oceanographic Literature Review, 10(42):892, 1995. D. G. Ainley, J. Russell, S. Jenouvrier, E. Woehler, P. O. Lyver, W. R. Fraser, and G. L. Kooyman. Antarctic penguin response to habitat change as earth’s troposphere reaches 2◦C above preindustrial levels. Ecological Monographs, 80:49– 66, 2010. URL http://www.esajournals.org/doi/pdf/10.1890/08-2289.1. H. Akaike. Information theory and an extension of the maximum likelihood princi- ple. In B. N. Petrov and F. Csaki, editors, Second International Symposium on Information Theory, pages 267–281. Academiai Kiado, Budapest, 1973. R. Altwegg, A. Roulin, M. Kestenholz, and L. Jenni. Demographic effects of extreme winter weather in the barn owl. Oecologia, 149(1):44–51, 2006. R. Ambrosini, N. Saino, D. Rubolini, and A. Møller. Higher degree-days at the time of breeding predict size of second clutches in the barn swallow. Climate Research, 50(1):43–50, 2011.

185 S. C. Amstrup, T. L. McDonald, and B. F. J. Manly. Handbook of Capture-Recapture Analysis. Princeton University Press, 2005. S. Ancona, I. Calixto-Albarr´an,and H. Drummond. Effect of El Ni˜noon the diet of a specialist seabird, Sula nebouxii, in the warm eastern tropical Pacific. Marine Ecology Progress Series, 462:261–271, 2012. D. R. Anderson, K. P. Burnham, and G. C. White. Problems in estimating age- specific survival rates from recovery data of birds ringed as young. Journal of Animal Ecology, 54:89–98, 1985. N. P. Ashmole. The regulation of numbers in tropical oceanic birds. Ibis, 103: 458–473, 1963. J. Austin, Oliver L. Notes on banding birds in Antarctica, and on the Ad´elie penguin colonies of the Ross Sea sector. Bird-Banding, 28(1):1–26, 1957. Australian Bureau of Meteorology. Climate statistics for Aus- tralian locations: definitions for temperature, 2007. URL http://www.bom.gov.au/climate/cdo/about/definitionstemp.shtml. Accessed: 5 December, 2014. Australian Bureau of Meteorology. Observation of air temperature, 2011. URL http://www.bom.gov.au/climate/cdo/about/airtemp-measure.shtml. Ac- cessed: 5 December, 2014. T. Ballerini, G. Tavecchia, S. Olmastroni, F. Pezzo, and S. Focardi. Nonlinear effects of winter sea ice on the survival probabilities of ad´eliepenguins. Oecologia, 161 (2):253–265, 2009. C. Barbraud and H. Weimerskirch. Emperor penguins and climate change. Nature, 411:183–186, 2001. C. Barbraud and H. Weimerskirch. Climate and density shape population dynamics of a marine top predator. Proceedings of the Royal Society of London B: Biological Sciences, 270(1529):2111–2116, 2003. C. Barbraud, P. Rivalan, P. Inchausti, M. Nevoux, V. Rolland, and H. Weimer- skirch. Contrasted demographic responses facing future climate change in south- ern ocean seabirds. Journal of Animal Ecology, 80(1):90–100, 2010. P. J. Barham, L. G. Underhill, R. J. M. Crawford, T. M. Leshoro, and D. A. Bolton. Impact of flipper banding on breeding success of African penguins Spheniscus dimersus at Robben Island: comparisons among silicone rubber bands, stainless- steel bands and no bands. African Journal of Marine Science, 30:595–602, 2008. R. J. Barker. Joint analysis of mark-recapture, resighting and ring-recovery data with age-dependence and marking-effect. Bird Study, 46 suppl.:82–90, 1999. R. V. Baudinette, P. Gill, and M. O’Driscoll. Energetics of the little penguin, Eudyptula minor: Temperature regulation, the calorigenic effect of food, and moulting. Australian Journal of Zoology, 34(1):35–45, 1986.

186 A. M. M. Baylis, B. Page, and S. D. Goldsworthy. Effect of seasonal changes in upwelling activity on the foraging locations of a wide-ranging central-place forager, the New Zealand fur seal. Canadian Journal of Zoology, 86(8):774–789, 2008. B. H. Becker, M. Peery, and S. R. Beissinger. Ocean climate and prey availability affect the trophic level and reproductive success of the marbled murrelet, an endangered seabird. Marine Ecology Progress Series, 329:267–279, 2007. M. Berlincourt and J. P. Arnould. Influence of environmental conditions on foraging behaviour and its consequences on reproductive performance in little penguins. Marine Biology, 167(2):1485–1501, 2015. M. Bertellotti, G. Blanco, G. R. Bortolotti, J. A. Donzar, M. G. Forero, and K. A. Hobson. Food resource utilisation by Magellanic penguins evaluated through stable isotope analysis: segregation by sex and age and influence of offspring quality. Marine Ecology Progress Series, 234:289–299, 2002. P. D. Boersma. The Gal´apagos penguin: a study of adaptation for life in an unpre- dictable environment. PhD thesis, Ohio State University, 1974. P. D. Boersma. Adaptation of Gal´apagospenguins for life in two different envi- ronments. In B. Stonehouse, editor, The Biology of Penguins, pages 101–114. Macmillan Press, 1975. P. D. Boersma. An ecological and behavioral study of the Gal´apagospenguins as an indicator of oceanographic conditions. The Living Bird, 15:43–93, 1977. P. D. Boersma. Breeding patterns of Gal´apagospenguins as an indicator of oceano- graphic conditions. Science, 200(4349):1481–1483, 1978. P. D. Boersma. Penguins as marine sentinels. BioScience, 58(7):597–607, 2008. P. D. Boersma and G. A. Rebstock. Flipper bands do not affect foraging-trip duration of Magellanic penguins. Journal of Field Ornithology, 80(4):408–418, 2009. P. D. Boersma and G. A. Rebstock. Effects of double bands on Magellanic penguins. Journal of Field Ornithology, 81(2):195–205, 2010. P. D. Boersma and G. A. Rebstock. Climate change increases reproductive failure in Magellanic penguins. PLoS ONE, 9(1):e85602, 2014. P. D. Boersma, N. T. Wheelwright, M. K. Nerini, and E. S. Wheelwright. The breeding biology of the fork-tailed storm-petrel Oceanodroma furcata. The Auk, 17:268–282, 1980. P. D. Boersma, G. A. Rebstock, E. Frere, and S. E. Moore. Following the fish: penguins and productivity in the South Atlantic. Ecological Monographs, 79(1): 59–76, 2009. R. Bonhomme. Bases and limits to using degree day units. European Journal of Agronomy, 13(1):1–10, 2000.

187 C. A. Bost and Y. Le Maho. Seabirds as bio-indicators of changing marine ecosys- tems: new perspectives. Acta Oecologia, 14:463–470, 1993. C. A. Bost, C. Cott´e,F. Bailleul, Y. Cherel, J. B. Charrassin, C. Guinet, D. G. Ainley, and H. Weimerskirch. The importance of oceanographic fronts to marine birds and mammals of the southern oceans. Journal of Marine Systems, 78(3): 363–376, 2009. L. W. Botsford, C. A. Lawrence, E. P. Dever, A. Hastings, and J. Largier. Effects of variable winds on biological productivity on continental shelves in coastal up- welling systems. Deep Sea Research Part II: Topical Studies in Oceanography, 53 (25):3116–3140, 2006. D. G. Boyce, M. R. Lewis, and B. Worm. Global phytoplankton decline over the past century. Nature, 466(7306):591–596, 2010. J. S. Bradley, R. D. Wooller, I. J. Skira, and D. L. Serventy. Age-dependent survival of breeding short-tailed shearwaters Puffinus tenuirostris. Journal of Animal Ecology,, 58:175–188, 1989. A. R. Breton, G. A. Fox, and J. W. Chardine. Survival of adult herring gulls Larus argentatus from a Lake Ontario colony over two decades of environmental change. Waterbirds, 31(1):15–23, 2008. S. P. Brooks, E. A. Catchpole, and B. J. Morgan. Bayesian animal survival estima- tion. Statistical Science, 15(4):357–376, 2000. C. Brownie, D. R. Anderson, K. P. Burnham, and D. S. Robson. Statistical inference from band recovery data: a handbook. US Fish and Wildlife Resource Publication 131, 1978. L. Bull. Fidelity and breeding success of the blue penguin Eudyptula minor on Matiu-Somes Island, Wellington, New Zealand. New Zealand Journal of Zoology, 27(4):291–298, 2000. K. P. Burnham. A theory for combined analysis of ring-recovery and recapture data. In J.-D. Lebreton and P. M. North, editors, Marked Individuals in the Study of Bird Population, pages 199–213. Birkh¨auserVerlag, Basel, 1993. K. P. Burnham and D. R. Anderson. Model Selection and Inference. A Practical Information-theoretic Approach. Springer, New York, 1998. K. P. Burnham and D. R. Anderson. Model Selection and Multimodel Inference. A Practical Information-theoretic Approach. Springer, New York, 2nd edition, 2002. doi:10.1007/b97636. K. P. Burnham, D. R. Anderson, G. C. White, C. Brownie, and K. H. Pollock. Design and Analysis Methods for Fish Survival Experiments based on Release- Recapture, Monograph 5. American Fisheries Society, Bethesda, Maryland, 1987. A. Butler, F. Althaus, D. Furlani, and K. Ridgway. Assessment of the conservation values of the Bonney upwelling area: a component of the Commonwealth Marine

188 Conservation Assessment Program 2002-2004: report to Environment Australia. CSIRO Marine Research Hobart, 2002. D. Cairns. Seabirds as indicators of marine food supplies. Biological Oceanography, 5(4):261–271, 1988. D. K. Cairns. The regulation of seabird colony size: a hinterland model. American Naturalist, 134:141–146, 1989. D. K. Cairns. Population regulation of seabird colonies. In Current ornithology, pages 37–61. Springer, 1992. B. Cannell, K. Pollock, S. Bradley, R. Wooller, W. Sherwin, and J. Sinclair. Aug- menting mark-recapture with beach counts to estimate the abundance of little penguins on Penguin Island, Western Australia. Wildlife Research, 38(6):491–500, 2011. B. Cannell, K. Campbell, L. Fitzgerald, J. Lewis, I. Baran, and N. Stephens. An- thropogenic trauma is the most prevalent cause of mortality in little penguins Eudyptula minor in Perth, Western Australia. Emu, 116(1):52–61, 2016. B. L. Cannell, L. E. Chambers, R. D. Wooller, and J. S. Bradley. Poorer breeding by little penguins near Perth, Western Australia is correlated with above average sea surface temperatures and a stronger Leeuwin Current. Marine and Freshwater Research, 63(10):914–925, 2012. E. A. Catchpole. Matlab: an environment for analyzing ring-recovery and recapture data. Journal of Applied Statistics,, 22:801–816, 1995. E. A. Catchpole, S. N. Freeman, B. J. T. Morgan, and M. P. Harris. Integrated recovery/recapture data analysis. Biometrics, 54:33–46, 1998. E. A. Catchpole, B. J. T. Morgan, S. N. Freeman, S. D. Albon, and T. N. Coul- son. Factors influencing Soay sheep survival. Technical Report UKC/IMS/99/17, University of Kent at Canterbury, England, 1999a. E. A. Catchpole, B. J. T. Morgan, S. N. Freeman, and W. J. Peach. Modelling the survival of British lapwings Vanellus vanellus using ring-recovery data and weather covariates. Bird Study, 46 suppl.:5–13, 1999b. E. A. Catchpole, B. J. Morgan, T. Coulson, S. N. Freeman, and S. D. Albon. Factors influencing Soay sheep survival. Journal of the Royal Statistical Society: Series C (Applied Statistics), 49(4):453–472, 2000. E. A. Catchpole, Y. Fan, B. J. T. Morgan, and T. N. Coulson. Sexual dimor- phism, survival and dispersal in red deer. Journal of Agricultural, Biological, and Environmental Statistics, 9:1–26, 2004. L. Chambers, L. Renwick, and P. Dann. Climate, fire and little penguins. In K. Richardson, W. Steffen, and D. Liverman, editors, Climate Change: Global Risks, Challenges and Decisions. Cambridge University Press, Cambridge, 2011a.

189 L. E. Chambers. Delayed breeding in little penguins – Evidence of climate change? Australian Meteorological Magazine, 53:13–19, 2004a. L. E. Chambers. The impact of climate on Little Penguin breeding success. Aus- tralian Bureau of Meteorology Research Centre, Research Report 100, 2004b. L. E. Chambers and P. Dann. An assessment of phenological change with ocean temperatures and interactions over the adult life cycle of a resident seabird. In review, 2015. L. E. Chambers, C. A. Devney, B. C. Congdon, N. Dunlop, E. J. Woehler, and P. Dann. Observed and predicted effects of climate on Australian seabirds. Emu, 111:235–251, 2011b. doi:10.1071/MU10033. L. E. Chambers, P. Dann, C. Deveney, N. Dunlop, and E. Woehler. Seabirds. Marine Climate Change in Australia: Impacts and Adaptation Responses 2012 Report Card, 2012:401–428, 2012. A. Chiaradia and I. C. Nisbet. Plasticity in parental provisioning and chick growth in little penguins Eudyptula minor in years of high and low breeding success. Ardea-Wageningen, 94(2):257, 2006. A. Chiaradia, M. G. Forero, K. A. Hobson, S. E. Swearer, F. Hume, L. Renwick, and P. Dann. Diet segregation between two colonies of little penguins Eudyptula minor in southeast Australia. Austral Ecology, 37(5):610–619, 2012. A. F. Chiaradia and K. R. Kerry. Daily nest attendance and breeding performance in the little penguin Eudyptula minor at Phillip Island, Australia. Marine Or- nithology, 27:13–20, 1999. R. Choquet, L. Rouan, and R. Pradel. Program E-SURGE: a software application for fitting multievent models. In Modeling Demographic Processes in Marked Populations, pages 845–865. Springer, 2009. A. Clarke, L. J. Holmes, and M. G. White. The annual cycle of temperature, chlorophyll and major nutrients at Signy Island, South Orkney Islands, 1969–82. Br. Antarct. Surv. Bull, 80:65–86, 1988. J. Clarke. Partitioning of foraging effort in Ad´eliepenguins provisioning chicks at B´echervaise Island, Antarctica. Polar Biology, 24(1):16–20, 2001. T. Clutton-Brock and B. C. Sheldon. Individuals and populations: the role of long- term, individual-based studies of animals in ecology and evolutionary biology. Trends in Ecology & Evolution, 25(10):562–573, 2010. T. H. Clutton-Brock, M. Major, and F. E. Guinness. Population regulation in male and female red deer. Journal of Animal Ecology, 54(3):831–846, 1985. ISSN 00218790. URL http://www.jstor.org/stable/4381. F. Colchero, O. R. Jones, and M. Rebke. Basta: an R package for Bayesian estima- tion of age-specific survival from incomplete mark–recapture/recovery data with covariates. Methods in Ecology and Evolution, 3(3):466–470, 2012.

190 M. Collins, J. M. Cullen, and P. Dann. Seasonal and annual foraging movements of little penguins from Phillip Island, Victoria. Wildlife Research, 26:705–721, 1999. doi:10.1071/WR98003. B. C. Congdon, C. A. Erwin, D. R. Peck, G. B. Baker, M. C. Double, and P. O’Neill. Vulnerability of seabirds on the Great Barrier Reef to climate change. The Great Barrier Reef Marine Park Authority, 2007. ISBN 9781876945619. E. G. Cooch and G. W. White. Program MARK: A gentle introduction, 2006. 5th edition. URL: http://www.phidot.org/software/mark/docs/book/. R. M. Cormack. Estimates of survival from the sighting of marked animals. Biometrika, 51:429–438, 1964. doi:10.1093/biomet/51.3-4.429. D. P. Costa. Reproductive and foraging energetics of high latitude penguins, alba- trosses and pinnipeds: implications for life history patterns. American Zoologist, 31(1):111–130, 1991. G. F. Cota, J. Wang, and J. C. Comiso. Transformation of global satellite chloro- phyll retrievals with a regionally tuned algorithm. Remote Sensing of Environ- ment, 90(3):373–377, 2004. C. Cott´e,Y.-H. Park, C. Guinet, and C.-A. Bost. Movements of foraging king penguins through marine mesoscale eddies. Proceedings of the Royal Society of London B: Biological Sciences, 274(1624):2385–2391, 2007. R. J. Crawford and B. M. Dyer. Responses by four seabird species to a fluctuating availability of Cape anchovy Engraulis capensis off South Africa. Ibis, 137(3): 329–339, 1995. R. J. M. Crawford, P. J. Barham, L. G. Underhill, L. J. Shannon, J. C. Coetzee, B. M. Dyer, T. M. Leshoro, and L. Upfold. The influence of food availability on breeding success of African penguins Spheniscus demersus at Robben Island, South Africa. Biological Conservation, 132(1):119–125, 2006. L. Crespin, M. P. Harris, J.-D. Lebreton, M. Frederiksen, and S. Wanless. Recruit- ment to a seabird population depends on environmental factors and on population size. Journal of Animal Ecology, 75:228–238, 2006. H. Q. P. Crick. The impact of climate change on birds. Ibis, 146:48–56, 2004. J. P. Croxall. Energy costs of incubation and moult in petrels and penguins. Journal of Animal Ecology, 51(1):177–194, 1982. J. P. Croxall, P. N. Trathan, and E. J. Murphy. Environmental change and Antarctic seabird populations. Science, 297(5586):1510–1514, 2002. S. Cubaynes, C. Lavergne, and O. Gimenez. Fitting animal survival models with temporal random effects. Environmental and Ecological Statistics, 21(3):599–610, 2014.

191 B. M. Culik, R. P. Wilson, and R. Bannasch. Flipper-bands on penguins: what is the cost of a life-long commitment? Marine Ecology Progress Series, 98:209–214, 1993. B. M. Culik, R. Bannasch, and R. P. Wilson. External devices on penguins: how important is shape? Marine Biology, 118(3):353–357, 1994. J. Cullen, T. Montague, and C. Hull. Food of little penguins Eudyptula minor in Victoria: Comparison of three localities between 1985 and 1988. Emu, 91(5): 318–341, 1992. J. M. Cullen, L. E. Chambers, P. C. Coutin, and P. Dann. Predicting onset and success of breeding in little penguins Eudyptula minor from ocean temperatures. Marine Ecology Progress Series, 378:269–278, 2009. doi:10.3354/meps07881. P. M. Cury, I. L. Boyd, S. Bonhommeau, T. Anker-Nilssen, R. J. Crawford, R. W. Furness, J. A. Mills, E. J. Murphy, H. Osterblom,¨ M. Paleczny, et al. Global seabird response to forage fish depletion — one-third for the birds. Science, 334 (6063):1703–1706, 2011. P. Dann. Distribution, population trends and factors influencing the population size of little penguins Eudyptula minor on Phillip Island, Victoria. Emu, 91(5): 263–272, 1992. doi:10.1071/MU9910263. P. Dann. The abundance, breeding distribution and nest sites of blue penguins in Otago, New Zealand. Notornis, 41(3), 1994. P. Dann. Little penguin Eudyptula minor. In P. G. Borboroglu and P. D. Boersma, editors, Penguins: Natural History and Conservation, pages 305–319. University of Washington Press, 2015. P. Dann and L. Chambers. Ecological effects of climate change on little penguins Eudyptula minor and the potential economic impact on tourism. Climate Re- search, 58:67–79, 2013. P. Dann and J. M. Cullen. Survival, patterns of reproduction and lifetime repro- ductive output in the little blue penguins (Eudyptula minor) on Phillip Island, Victoria, Australia. In L. S. Davis and J. T. Darby, editors, Penguin Biology, pages 63–84. Academic Press, San Diego, 1990. P. Dann and F. I. Norman. Population regulation in little penguins, Eudyptula minor: the role of intraspecific competition for nesting sites and food during breeding. Emu, 106:289–296, 2006. P. Dann, J. Cullen, R. Thoday, and R. Jessop. Movements and patterns of mortality at sea of little penguins Eudyptula minor from Phillip Island, Victoria. Emu, 91 (5):278–286, 1992. P. Dann, J. M. Cullen, and R. Jessop. Cost of reproduction in little penguins. In P. Dann, F. I. Norman, and P. N. Reilly, editors, The Penguins: Ecology and Management, pages 39–55. Surrey Beatty, Sydney, 1995.

192 P. Dann, F. I. Norman, J. M. Cullen, F. J. Neira, and A. F. Chiaradia. Mortality and breeding failure of little penguins, Eudyptula minor, in Victoria, 1995–96, follow- ing a widespread mortality of pilchard, Sardinops sagax. Marine and Freshwater Research, 51(4):355–362, 2000. P. Dann, M. Carron, B. Chambers, L. Chambers, T. Dornom, A. McLaughlin, B. Sharp, M. E. Talmage, R. Thoday, and S. Unthank. Longevity in little pen- guins Eudyptula minor. Marine Ornithology, 33:71–72, 2005. P. Dann, L. A. Sidhu, R. Jessop, L. Renwick, M. Healy, P. Collins, G. B. Baker, and E. Catchpole. Effects of flipper bands and injected transponders on the survival of adult little penguins Eudyptula minor. Ibis, 156:73–83, 2014. L. S. Davis. Creching behaviour of Ad´eliepenguin chicks Pygoscelis adeliae. New Zealand Journal of Zoology, 9(2):279–285, 1982. S. E. Davis, R. G. Nager, and R. W. Furness. Food availability affects adult survival as well as breeding success of parasitic jaegers. Ecology, 86(4):1047–1056, 2005. M. B. Decker, J. G. L. Hunt, and J. G. V. Byrd. The relationships among sea–surface temperature, the abundance of juvenile walleye pollock Theragra chalcogramma, and the reproductive performance and diets of seabirds at the Pribilof Islands, southeastern Bering sea. In R. J. Beamish, editor, Climate Change and Northern Fish Populations, pages 425–437. National Research Council of Canada, Ottawa, Canada, 1995. E. S. Deevey, Jr. Life tables for natural populations of animals. The Quarterly Review of Biology, 22(4):283–314, 1947. N. Dehnhard, K. Ludynia, M. Poisbleau, L. Demongin, and P. Quillfeldt. Good days, bad days: wind as a driver of foraging success in a flightless seabird, the southern rockhopper penguin. PLoS One, 8(11):e79487, 2013. N. Dehnhard, M. Eens, L. Demongin, P. Quillfeldt, and M. Poisbleau. Individual consistency and phenotypic plasticity in rockhopper penguins: female but not male body mass links environmental conditions to reproductive investment. PloS One, 10(6):e0128776, 2015a. N. Dehnhard, M. Eens, L. Demongin, P. Quillfeldt, D. Suri, and M. Poisbleau. Limited individual phenotypic plasticity in the timing of and investment into egg laying in southern rockhopper penguins under climate change. Inter-Research Marine Ecology Progress Series, 524:269–281, 2015b. S. C. Doney, M. Ruckelshaus, J. E. Duffy, J. P. Barry, F. Chan, C. A. English, H. M. Galindo, J. M. Grebmeier, A. B. Hollowed, N. Knowlton, et al. Climate change impacts on marine ecosystems. Marine Science, 4, 2012. K. M. Dugger, G. Ballard, D. G. Ainley, and K. J. Barton. Effects of flipper bands on foraging behaviour and survival of Ad´eliepenguins Pygoscelis adeliae. The Auk, 123(3):858–869, 2006.

193 G. Dunnet and J. C. Ollason. The estimation of survival rate in the fulmar, Ful- marus glacialis. The Journal of Animal Ecology, pages 507–520, 1978. J. A. Dupuis. Bayesian estimation of movement and survival probabilities from capture-recapture data. Biometrika, 82(4):761–772, 1995. J. M. Durant, T. Anker-Nilssen, and N. C. Stenseth. Trophic interactions under climate fluctuations: the Atlantic puffin as an example. Proceedings of the Royal Society of London B: Biological Sciences, 270(1523):1461–1466, 2003. Ernst and Young. Economic Impact of Phillip Island Nature Parks. Phillip Island Nature Parks, 2012. P. M. Fallow, A. Chiaradia, Y. Ropert-Coudert, A. Kato, and R. D. Reina. Flip- per bands modify the short-term diving behavior of little penguins. Journal of Wildlife Management, 73(8):1348–1354, 2009. W. T. Federer. Experimental Design, Theory and Application. Macmillan, 1955. A.´ M. Felic´ısimo, J. Mu˜noz,and J. Gonz´alez-Solis. Ocean surface winds drive dynamics of transoceanic aerial movements. PLoS One, 3(8):e2928, 2008. J. Forcada and P. N. Trathan. Penguin responses to climate change in the Southern Ocean. Global Change Biology, 15(7):1618–1630, 2009. M. Fortescue. Temporal and spatial variation in breeding success of the Little Penguin Eudyptula minor on the east coast of Australia. Marine Ornithology, 27:21–28, 1999. D. C. Franklin, G. Ehmke, J. VanDerWal, and S. T. Garnett. The exposure of Australian birds to climate change. In S. T. Garnett and D. C. Franklin, edi- tors, Climate Change Adaptation Plan for Australian Birds, pages 7–26. CSIRO Publishing, Victoria, Australia, 2014. M. Frederiksen, F. Daunt, M. P. Harris, and S. Wanless. The demographic impact of extreme events: stochastic weather drives survival and population dynamics in a long-lived seabird. Journal of Animal Ecology, 77(5):1020–1029, 2008. M. Frederiksen, J.-D. Lebreton, R. Pradel, R. Choquet, and O. Gimenez. Review: Identifying links between vital rates and environment: a toolbox for the applied ecologist. Journal of Applied Ecology, 51(1):71–81, 2014. S. N. Freeman and B. J. T. Morgan. A modelling strategy for recovery data from birds ringed as nestlings. Biometrics, 48:217–236, 1992. E. Frere, P. Gandini, and D. Boersma. The breeding ecology of Magellanic pen- guins at Cabo V´ırgenes, Argentina: what factors determine reproductive success? Colonial Waterbirds, pages 205–210, 1998. G. Froget, M. Gauthier-Clerc, Y. Le Maho, and Y. Handrich. Is penguin banding harmless? Polar Biology, 20:409–413, 1998.

194 P. G. H. Frost, W. R. Siegfried, and A. E. Burger. Behavioural adaptations of the jackass penguin, Spheniscus demersus to a hot, arid environment. Journal of Zoology, 179(2):165–187, 1976. R. W. Furness. Impacts of fisheries on seabird communities. Scientia Marina, 67 (S2):33–45, 2003. R. W. Furness and T. R. Birkhead. Seabird colony distributions suggest competition for food supplies during the breeding season. Nature, 311:655–666, 1984. R. W. Furness and K. C. Camphuysen. Seabirds as monitors of the marine environment. ICES Journal of Marine Science: Journal du Conseil, 54(4):726–737, 1997. doi: 10.1006/jmsc.1997.0243. URL http://icesjms.oxfordjournals.org/content/54/4/726.abstract. R. Gales. Breeding seasons and double brooding of the little penguin Eudyptula minor in New Zealand. Emu, 85:127–130, 1985. R. Gales and B. Green. The annual energetics cycle of little penguins Eudyptula minor. Ecology, 71(6):2297–2312, 1990. R. Gales, B. Green, and C. Stahel. The energetics of free-living little penguins Eudyptula minor (Spheniscidae), during moult. Australian Journal of Zoology, 36(2):159–167, 1988. R. Gales, C. Williams, and D. Ritz. Foraging behaviour of the little penguin, Eudyptula minor: initial results and assessment of instrument effect. Journal of Zoology, 220(1):61–85, 1990. L. B. Ganendran. The Effect of Wind on the Survival of Little Penguins. Master’s thesis, The University of New South Wales, Canberra, 2011. L. B. Ganendran, L. A. Sidhu, E. A. Catchpole, P. Dann, and L. E. Chambers. The effect of directional wind components on survival of little penguins Eudyptula minor. ANZIAM J (E), 52:C1012–C1030, 2011. L. B. Ganendran, L. A. Sidhu, E. A. Catchpole, L. E. Chambers, and P. Dann. Effects of ambient air temperature, humidity and rainfall on survival of adult little penguins Eudyptula minor in southeastern Australia. International Journal of Biometeorology, 60(8):1237–1245, 2016. T. Garrison. Oceanography—An Invitation to Marine Science. Brooks/Cole- Thomson Learning, Belmont, 5th edition, 2005. A. J. Gaston, J. M. Hipfner, and D. Campbell. Heat and mosquitoes cause breeding failures and adult mortality in an arctic-nesting seabird. Ibis, 144(2):185–191, 2002. M. Gauthier-Clerc, J.-P. Gendner, C. A. Ribic, W. R. Fraser, E. J. Woehler, S. Descamps, C. Gilly, C. Le Bohec, and Y. Le Maho. Long-term effects of flipper bands on penguins. Proceedings of the Royal Society of London B (Supplement), Biology Letters, 271:423–426, 2004.

195 C. F. Gibbs. Oceanography of Bass Strait: Implications for the food supply of little penguins Eudyptula minor. Emu, 91(5):395–401, 1992. P. C. Gill, M. G. Morrice, B. Page, R. Pirzl, A. H. Levings, and M. Coyne. Blue whale habitat selection and within-season distribution in a regional upwelling system off southern Australia. Marine Ecology Progress Series, 421:243–263, 2011. V. A. Gill, S. A. Hatch, and R. B. Lanctot. Sensitivity of breeding parameters to food supply in black-legged Kittiwakes Rissa tridactyla. Ibis, 144(2):268–283, 2002. O. Gimenez, R. Choquet, and J.-D. Lebreton. Parameter redundancy in multistate capture-recapture models. Biometrical Journal, 45(6):704–722, 2003. O. Gimenez, V. Rossi, R. Choquet, C. Dehais, B. Doris, H. Varella, J.-P. Vila, and R. Pradel. State-space modelling of data on marked individuals. Ecological Modelling, 206(3):431–438, 2007. R. Gradinger. Climate change and biological oceanography of the Arctic Ocean. Philosophical Transactions of the Royal Society of London A: Mathematical, Physical and Engineering Sciences, 352(1699):277–286, 1995. W. W. Gregg, N. W. Casey, and C. R. McClain. Recent trends in global ocean chlorophyll. Geophysical Research Letters, 32(3), 2005. D. Gr´emillet,T. Boulinier, et al. Spatial ecology and conservation of seabirds facing global climate change: a review. Marine Ecology Progress Series, 391:121–137, 2009. J. Griffin. Penguins feel the heat. Quest, 1(3):17, 2005. V. Grosbois and P. M. Thompson. North Atlantic climate variation influences survival in adult fulmars. Oikos, 109(2):273–290, 2005. V. Grosbois, P.-Y. Henry, J. Blondel, P. Perret, J.-D. Lebreton, D. W. Thomas, and M. M. Lambrechts. Climate impacts on Mediterranean blue tit survival: an investigation across seasons and spatial scales. Global Change Biology, 12(12): 2235–2249, 2006. V. Grosbois, O. Gimenez, J. M. Gaillard, R. Pradel, C. Barbraud, J. Clobert, A. P. Moller, and H. Weimerskirch. Assessing the impact of climate variation on survival in vertebrate populations. Biological Reviews, 83(3):357–399, 2008. R. Groscolas. Study of moult fasting followed by an experimental forced fasting in the emperor penguin Aptenodytes forsteri: Relationship between feather growth, body weight loss, body temperature and plasma fuel levels. Comparative Bio- chemistry and Physiology Part A: Physiology, 61(2):287–295, 1978. C. Guinet, O. Chastel, M. Koudil, J. P. Durbec, and P. Jouventin. Effects of warm sea surface temperature anomalies on the blue petrel at the Kerguelen Islands.

196 Proceedings of the Royal Society of London B: Biological Sciences, 265(1400): 1001–1006, 1998. S. E. Gutowsky, Y. Tremblay, M. A. Kappes, E. N. Flint, J. Klavitter, L. Laniawe, D. P. Costa, M. B. Naughton, M. D. Romano, and S. A. Shaffer. Divergent post- breeding distribution and habitat associations of fledgling and adult black-footed albatrosses Phoebastria nigripes in the North Pacific. Ibis, 156(1):60–72, 2014. A. S. Hadi and R. F. Ling. Some cautionary notes on the use of principal components regression. The American Statistician, 52(1):15–19, 1998. A. Hald. A History of Probability and Statistics and their Applications before 1750. John Wiley & Sons, New York, 2003. B. S. Halpern, S. Walbridge, K. A. Selkoe, C. V. Kappel, F. Micheli, C. D’Agrosa, J. F. Bruno, K. S. Casey, C. Ebert, H. E. Fox, et al. A global map of human impact on marine ecosystems. Science, 319(5865):948–952, 2008. S. L. Hampton, P. G. Ryan, and L. G. Underhill. The effect of flipper banding on the breeding success of African penguins Spheniscus demersus at Boulders Beach, South Africa. Ostrich, 80(2):77–80, 2009. G. Haramis, J. Nichols, K. Pollock, and J. Hines. The relationship between body mass and survival of wintering canvasbacks. The Auk, 103:506–514, 1986. A. Harding, R. Paredes, R. Suryan, D. Roby, D. Irons, R. Orben, H. Renner, R. Young, C. Barger, I. Dorresteijn, and A. Kitaysky. Does location really mat- ter? An inter-colony comparison of seabirds breeding at varying distances from productive oceanographic features in the Bering Sea. Deep Sea Research Part II: Topical Studies in Oceanography, 94:178–191, 2013. A. M. Harding, J. Welcker, H. Steen, K. C. Hamer, A. S. Kitaysky, J. Fort, S. L. Talbot, L. A. Cornick, N. J. Karnovsky, G. W. Gabrielsen, and D. Gr´emillet. Adverse foraging conditions may impact body mass and survival of a high Arctic seabird. Oecologia, 167(1):49–59, 2011. K. E. Harrigan. Causes of mortality of little penguins Eudyptula minor in Victoria. Emu, 91:273–277, 1992. M. Harris and K. Bode. Populations of little penguins, short-tailed shearwaters and other seabirds on Phillip Island, Victoria, 1978. Emu, 81(1):20–28, 1981. M. P. Harris, S. N. Freeman, S. Wanless, B. J. T. Morgan, and C. V. Wernham. Factors influencing the survival of puffins Fratercula arctica at a North Sea colony over a 20-year period. Journal of Avian Biology, 28:287–295, 1997. M. P. Harris, T. Anker-Nilssen, R. H. McCleery, K. E. Erikstad, D. N. Shaw, and V. Grosbois. Effect of wintering area and climate on the survival of adult Atlantic puffins Fratercula arctica in the eastern Atlantic. Marine Ecology Progress Series, 297(283–296), 2005.

197 M. P. Harris, M. Frederiksen, and S. Wanless. Within-and between-year variation in the juvenile survival of common guillemots Uria aalge. Ibis, 149(3):472–481, 2007. S. Heber, K.-J. Wilson, and L. Molles. Breeding biology and breeding success of the blue penguin (Eudyptula minor) on the west coast of New Zealand’s South Island. New Zealand Journal of Zoology, 35(1):63–71, 2008. G. Hemery, F. D’Amico, I. Castege, B. Dupont, J. D’Elbee, Y. Lalanne, and C. Mouches. Detecting the impact of oceano-climatic changes on marine ecosys- tems using a multivariate index: the case of the Bay of Biscay (North Atlantic- European Ocean). Global Change Biology, 14(1):27–38, 2008. M. A. Hindell, M.-A. Lea, and C. L. Hull. The effects of flipper bands on adult survival rate and reproduction in the royal penguin, Eudyptes schlegeli. Ibis, 138 (3):557–560, 1996. J. M. Hipfner. Matches and mismatches: ocean climate, prey phenology and breed- ing success in a zooplanktivorous seabird. Marine Ecology Progress Series, 368: 295–304, 2008. A. J. Hobday, L. E. Chambers, and J. P. Arnould. Methods to prioritise adaptation options for iconic seabirds and marine mammals impacted by climate change. In J. Palutikof, S. Boulter, J. Barnett, and D. Rissik, editors, Applied Studies in Climate Adaptation, pages 77–94. 2014. D. K. Hobday. Abundance and distribution of pilchard and Australian anchovy as prey species for the little penguin Eudyptula minor at Phillip Island, Victoria. Emu, 91(5):342–354, 1992. J. Hodder and M. R. Graybill. Reproduction and survival of seabirds in Oregon during the 1982–1983 El Ni˜no. The Condor, 87(4):535–541, 1985. C. Horswill, J. Matthiopoulos, J. A. Green, M. P. Meredith, J. Forcada, H. Peat, M. Preston, P. N. Trathan, and N. Ratcliffe. Survival in macaroni penguins and the relative importance of different drivers: individual traits, predation pressure and environmental variability. Journal of Animal Ecology, 83(5):1057–1067, 2014. A. J. Hoskins, P. Dann, Y. Ropert-Coudert, A. Kato, A. Chiaradia, D. P. Costa, and J. P. Y. Arnould. Foraging behaviour and habitat selection at sea in little penguins Eudyptula minor during the chick-guard stage of breeding. Marine Ecology Progress Series, 366:293–303, 2008. J. E. Hovinen, J. Welcker, S. Descamps, H. Strøm, K. Jerstad, J. Berge, and H. Steen. Climate warming decreases the survival of the little auk (Alle alle), a high Arctic avian predator. Ecology and Evolution, 4(15):3127–3138, 2014. K. D. Hyrenbach and R. R. Veit. Ocean warming and seabird communities of the southern California Current system (1987-98): response at multiple temporal

198 scales. Deep Sea Research Part II: Topical Studies in Oceanography, 50(14):2537– 2565, 2003. P. Inchausti, C. Guinet, M. Koudil, J.-P. Durbec, C. Barbraud, H. Weimerskirch, Y. Cherel, and P. Jouventin. Inter-annual variability in the breeding performance of seabirds in relation to oceanographic anomalies that affect the Crozet and the Kerguelen sectors of the Southern Ocean. Journal of Avian Biology, 34(2):170– 176, 2003. S. Jackson and R. P. Wilson. The potential costs of flipper-bands to penguins. Functional Ecology, 16:141–148, 2002. S. Jenouvrier. Impacts of climate change on avian populations. Global Change Biology, 19(7):2036–2057, 2013. S. Jenouvrier, C. Barbraud, and H. Weimerskirch. Long-term contrasted responses to climate of two antarctic seabird species. Ecology, 86(11):2889–2903, 2005a. S. Jenouvrier, H. Weimerskirch, C. Barbraud, Y. H. Park, and B. Cazelles. Evi- dence of a shift in the cyclicity of Antarctic seabird dynamics linked to climate. Proceedings of the Royal Society of London B, 272(1566):887–895, 2005b. S. Jenouvrier, C. P´eron,and H. Weimerskirch. Extreme climate events and individ- ual heterogeneity shape life-history traits and population dynamics. Ecological Monographs, 85(4):605–624, 2015. E. Johannesen, L. Perriman, and H. Steen. The effect of breeding success on nest and colony fidelity in the little penguin (Eudyptula minor) in Otago, New Zealand. Emu, 102:241–247, 2002a. E. Johannesen, H. Steen, and L. Perriman. Seasonal variation in survival, weights, and population counts of blue penguins Eudyptula minor in Otago, New Zealand. New Zealand Journal of Zoology, 29(3):213–219, 2002b. E. Johannesen, D. Houston, and J. Russell. Increased survival and breeding per- formance of double breeders in little penguins Eudyptula minor, New Zealand: evidence for individual bird quality? Journal of Avian Biology, 34:198–210, 2003. I. T. Jolliffe. A note on the use of principal components in regression. Applied Statistics, pages 300–303, 1982. I. T. Jolliffe. Principal component analysis: a beginner’s guide - i. introduction and application. Weather, 45(10):375–382, 1990. I. T. Jolliffe. Principal component analysis: a beginner’s guide - ii. pitfalls, myths and extensions. Weather, 48(8):246–253, 1993. G. M. Jolly. Explicit estimates from capture-recapture data with both death and immigration-stochastic models. Biometrika, 52:225–247, 1965. doi:10.2307/2333826.

199 D. A. Jones, W. Wang, and R. Fawcett. High-quality spatial climate data-sets for Australia. Australian Meteorological and Oceanographic Journal, 58(4):233–248, 2009. I. L. Jones, F. M. Hunter, and G. J. Robertson. Annual adult survival of least auklets (Aves, Alcidae) varies with large-scale climatic conditions of the North Pacific Ocean. Oecologia, 133(1):38–44, 2002. J. K¨ampf.Phytoplankton blooms on the western shelf of Tasmania: evidence of a highly productive ecosystem. Ocean Science, 11(1):1–11, 2015. doi: 10.5194/os- 11-1-2015. URL http://www.ocean-sci.net/11/1/2015/. J. K¨ampf, M. Doubell, D. Griffin, R. L. Matthews, and T. M. Ward. Evidence of a large seasonal coastal upwelling system along the southern shelf of Australia. Geophysical Research Letters, 31:L09310, 2004. K. B. Karnauskas, S. Jenouvrier, C. W. Brown, and R. Murtugudde. Strong sea surface cooling in the eastern equatorial Pacific and implications for Gal´apagos penguin conservation. Geophysical Research Letters, 42(15):6432–6437, 2015. P. Kelly, L. Clementson, and V. Lyne. Decadal and seasonal changes in temperature, salinity, nitrate, and chlorophyll in inshore and offshore waters along southeast Australia. Journal of Geophysical Research: Oceans, 120(6):4226–4244, 2015. A. S. Kitaysky, J. F. Piatt, S. A. Hatch, E. V. Kitaiskaia, Z. M. Benowitz-Fredericks, M. T. Shultz, and J. C. Wingfield. Food availability and population processes: severity of nutritional stress during reproduction predicts survival of long-lived seabirds. Functional Ecology, 24(3):625–637, 2010. N. Klomp, C. Meathrel, B. Wienecke, and R. Wooller. Surface nesting by little penguins on Penguin Island, Western Australia. Emu, 91(3):190–193, 1991. C. Knight and T. Rogers. Factors influencing fledgling production in little penguins Eudyptula minor. Wildlife Research, 31(3):339–344, 2004. D. Lack. The age of the blackbird. British Birds, 36:166–175, 1943. D. Lack. Breeding seasons in the Galapagos. Ibis, 92(2):268–278, 1950. D. Lack. The Natural Regulation of Animal Numbers. Clarendon Press, Oxford, 1954. D. L. Lack. Ecological adaptations for breeding in birds. 1968. K. H. Lakhani and I. Newton. Estimating age-specific bird surival rates from ring recoveries – can it be done? Journal of Animal Ecology, 52:83–91, 1983. C. Le Bohec, M. Gauthier-Clerc, and Y. Le Maho. The adaptive significance of cr´eches in the king penguin. Animal Behaviour, 70(3):527–538, 2005. C. Le Bohec, J. Durant, M. Gauthier-Clerc, N. Stenseth, Y.-H. Park, R. Pradel, D. Gr´emillet,J.-P. Gendner, and Y. Le Maho. King penguin population threat- ened by Southern Ocean warming. Proceedings of the National Academy of Sci- ences, 105(7):2493–2497, 2008.

200 C. Le Bohec, J. D. Whittington, and Y. Le Maho. Polar monitoring: seabirds as sentinels of marine ecosystems. In Adaptation and Evolution in Marine Environ- ments, Volume 2, pages 205–230. Springer, 2013. Y. Le Maho, C. Saraux, J. M. Durant, V. A. Viblanc, M. Gauthier-Clerc, N. G. Yoccoz, N. C. Stenseth, and C. Le Bohec. An ethical issue in biodiversity science: The monitoring of penguins with flipper bands. Comptes Rendus Biologies, 334 (5):378–384, 2011. M. Le Vaillant, Y. Ropert-Coudert, Y. Le Maho, and C. Le Bohec. Individual parameters shape foraging activity in breeding king penguins. Behavioral Ecology, page arv146, 2015. J.-D. Lebreton, K. P. Burnham, J. Clobert, and D. R. Anderson. Modeling survival and testing biological hypotheses using marked animals: a unified approach with case studies. Ecological Monographs, 62:67–118, 1992. J.-D. Lebreton, R. Pradel, and J. Clobert. The statistical analysis of survival in animal populations. Trends in Ecology & Evolution, 8(3):91–95, 1993. N. Lecomte, G. Kuntz, N. Lambert, J.-P. Gendner, Y. Handrich, Y. Le Maho, and C.-A. Bost. Alloparental feeding in the king penguin. Animal Behaviour, 71(2): 457–462, 2006. A. Lescro¨eland C.-A. Bost. Foraging under contrasting oceanographic conditions: the gentoo penguin at Kerguelen Archipelago. Marine Ecology Progress Series, 302:245–261, 2005. A. Lescro¨el,K. M. Dugger, G. Ballard, and D. G. Ainley. Effects of individual quality, reproductive success and environmental variability on survival of a long- lived seabird. Journal of Animal Ecology, 78(4):798–806, 2009. M. Lettink and D. P. Armstrong. An introduction to using mark-recapture anal- ysis for monitoring threatened species. Technical Report 28A, Department of Conservation New Zealand, 2003. R. Lewis. Seasonal upwelling along the south-eastern coastline of South Australia. Marine and Freshwater Research, 32(6):843–854, 1981. S. Lewis, S. Wanless, D. A. Elston, M. D. Schultz, E. Mackley, M. Du Toit, J. G. Underhill, and M. P. Harris. Determinants of quality in a long-lived colonial species. Journal of Animal Ecology, 75(6):1304–1312, 2006. S. Lewis, R. A. Phillips, S. J. Burthe, S. Wanless, and F. Daunt. Contrasting re- sponses of male and female foraging effort to year-round wind conditions. Journal of Animal Ecology, 84(6):1490–1496, 2015. M. S. Lindberg. A review of designs for capture-mark-recapture studies in discrete time. Journal of Ornithology, 152(2):355–370, 2012.

201 H. J. Lynch, R. Naveen, P. N. Trathan, and W. F. Fagan. Spatially integrated assessment reveals widespread changes in penguin populations on the Antarctic Peninsula. Ecology, 93(6):1367–1377, 2012. B. F. J. Manly. The determination of key factors from life table data. Oecologia, 31(1):111–117, 1977. B. F. J. Manly, T. L. McDonald, and S. C. Amstrup. Introduction to the handbook. In S. C. Amstrup, T. L. McDonald, and B. F. J. Manly, editors, Handbook of Capture-recapture Analysis. Princeton University Press, Princeton, 2005. K. Mann. Physical oceanography, food chains, and fish stocks: a review. ICES Journal of Marine Science: Journal du Conseil, 50(2):105–119, 1993. K. Mann and J. Lazier. Dynamics of Marine Ecosystems: Biological-Physical In- teractions in the Oceans. Blackwell Publishing, Oxford, 3rd edition, 2006. K. Marchetti and T. Price. Differences in the foraging of juvenile and adult birds: the importance of developmental constraints. Biological Reviews, 64(1):51–70, 1989. G. P. McCabe. Principal variables. Technometrics, 26(2):137–144, 1984. P. McCullagh and J. A. Nelder. Generalized Linear Models. Chapman and Hall, London, 2nd edition, 1989. C. McCutcheon, P. Dann, M. Salton, L. Renwick, A. J. Hoskins, A. M. Gormley, and J. P. Y. Arnould. The foraging range of little penguins Eudyptula minor during winter. Emu, 111:321–329, 2011. doi: 10.1071/MU10078. K. L. McInnes and G. D. Hubbert. A numerical modelling study of storm surges in Bass Strait. Australian Meteorological Magazine, 52(3), 2003. A. E. McKechnie and B. O. Wolf. Climate change increases the likelihood of catas- trophic avian mortality events during extreme heat waves. Biology Letters, 6: 253–256, 2010. D. J. McLeod, A. J. Hobday, J. M. Lyle, and D. C. Welsford. A prey-related shift in the abundance of small pelagic fish in eastern Tasmania? ICES Journal of Marine Science: Journal du Conseil, 69(6):953–960, 2012. M. Messi´eand M.-H. Radenac. Seasonal variability of the surface chlorophyll in the western tropical pacific from SeaWiFS data. Deep Sea Research Part I: Oceanographic Research Papers, 53(10):1581–1600, 2006. M. J. Mickelson, P. Dann, and J. M. Cullen. Sea temperature in Bass Strait and breeding success of the little penguins Eudyptula minor at Phillip Island, south- eastern Australia. Emu, 91(5):355–368, 1992. doi:10.1071/MU9910355. T. Micol and P. Jouventin. Long-term population trends in seven Antarctic seabirds at Pointe G´eologie(Terre Ad´elie):Human impact compared with environmental change. Polar Biology, 24(3):175–185, 2001.

202 J. A. Mills, J. W. Yarrall, J. M. Bradford-Grieve, M. J. Uddstrom, J. A. Renwick, and J. Meril¨a. The impact of climate fluctuation on food availability and repro- ductive performance of the planktivorous red-billed gull Larus novaehollandiae scopulinus. Journal of Animal Ecology, 77(6):1129–1142, 2008. P. I. Mitchell, S. F. Newton, N. Ratcliffe, and T. E. Dunn. Seabird populations of Britain and Ireland: resultsof the Seabird 2000 census (1998–2002). T and A.D. Poyser, London, 2004. P. Monaghan. Relevance of the behaviour of seabirds to the conservation of marine environments. Oikos, 77(2):227–237, 1996. ISSN 00301299. URL http://www.jstor.org/stable/3546061. P. Monaghan, J. Uttley, M. Burns, C. Thaine, and J. Blackwood. The relationship between food supply, reproductive effort and breeding success in Arctic terns Sterna paradisaea. The Journal of Animal Ecology, pages 261–274, 1989. T. Montague and J. M. Cullen. The diet of the little penguin Eudyptula minor at Phillip Island, Victoria. Emu, 88:138–149, 1987. I. Montero-Serra, M. Edwards, and M. J. Genner. Warming shelf seas drive the subtropicalization of European pelagic fish communities. Global Change Biology, 21(1):144–153, 2015. W. A. Montevecchi and R. A. Myers. Prey harvests of seabirds reflect pelagic fish and squid abundance on multiple spatial and temporal scales. Marine Ecology Progress Series, 117:1–9, 1995. J. Moreno, A. De Leon, J. Fargallo, and E. Moreno. Breeding time, health and immune response in the chinstrap penguin Pygoscelis antarctica. Oecologia, 115 (3):312–319, 1998. K. W. Morrison, N. C. Morrison, R. M. Buchheit, R. Dunn, P. F. Battley, and D. R. Thompson. The canalized parental roles of a Eudyptes penguin constrain provisioning and growth of chicks during nutritional stress. Behavioral Ecology and Sociobiology, 70(4):467–479, 2016. S. Nakagawa and I. C. Cuthill. Effect size, confidence interval and statistical signif- icance: a practical guide for biologists. Biological Reviews, 82(4):591–605, 2007. D. C. Nel and P. Whittington, editors. Rehabilitation of Oiled African Penguins: a Conservation Success Story. BirdLife South Africa, 2003. M. Nevoux and C. Barbraud. Relationships between sea ice concentration, sea surface temperature and demographic traits of thin-billed prions. Polar Biology, 29(6):445–453, 2006. A.-E. Nieblas, B. M. Sloyan, A. J. Hobday, R. Coleman, and A. J. Richardson. Variability of biological production in low wind-forced regional upwelling systems: A case study off southeastern Australia. Limnology and Oceanography, 54(5): 1548–1558, 2009.

203 I. C. T. Nisbet. Detecting and measuring senescence in wild birds: experience with long-lived seabirds. Experimental Gerontology, 36(4–6):833–843, 2001. I. C. T. Nisbet and P. Dann. Reproductive performance of little penguins in relation to year, age, pair-bond duration, breeding date and individual quality. Journal of Avian Biology, 40:296–308, 2009. doi:10.1111/j.1600-048X.2008.04563.x. M. Numata, L. S. Davis, and M. Renner. Prolonged foraging trips and egg desertion in little penguins Eudyptula minor. New Zealand Journal of Zoology, 27(4):277– 289, 2000. C. S. Oedekoven, D. G. Ainley, and L. B. Spear. Variable responses of seabirds to change in marine climate: California Current, 1985-1994. Marine Ecology Progress Series, 212:265–281, 2001. A. M. Olsen and S. A. Shepherd. Historic drift bottle experiments show reversing surface water masses in western Bass Strait waters: implications for lobster larval dispersal. Transactions of the Royal Society of South Australia, 130(1):113–122, 2006. S. A. Oswald and J. M. Arnold. Direct impacts of climatic warming on heat stress in endothermic species: seabirds as bioindicators of changing thermoregulatory constraints. Integrative Zoology, 7(2):121–136, 2012. R. Paredes, C. B. Zavalaga, D. J. Boness, and M. du Plessis. Patterns of egg laying and breeding success in Humboldt penguins Spheniscus humboldti at Punta San Juan, Peru. The Auk, 119(1):244–250, 2002. W. J. Peach, G. M. Siriwardena, and R. D. Gregory. Long-term changes in over- winter survival rates explain the decline of reed buntings Emberiza schoeniclus in Britain. Journal of Applied Ecology, 36(5):798–811, 1999. A. C. Perdeck, M. E. Visser, and J. H. Van Balen. Great tit Parus major survival and the beech-crop. Ardea, 88:99–106, 2000. L. Perriman and H. Steen. Blue penguin Eudyptula minor nest distribution and breeding success on Otago Peninsula, 1992 to 1998. New Zealand Journal of Zoology, 27(4):269–275, 2000. L. Perriman, D. Houston, H. Steen, and E. Johannesen. Climate fluctuation effects on breeding of blue penguins (Eudyptula minor). New Zealand Journal of Zoology, 27:261–267, 2000. S. L. Petersen, G. M. Branch, D. G. Ainley, P. D. Boersma, J. Cooper, and E. J. Woehler. Is flipper banding of penguins a problem? Marine Ornithology, 33: 75–79, 2006. A. J. Peucker, P. Dann, and C. P. Burridge. Range-wide phylogeography of the little penguin Eudyptula minor: Evidence of long-distance dispersal. The Auk, 126(2):397–408, 2009.

204 J. F. Piatt. The aggregative response of common murres and Atlantic puffins to schools of capelin. Studies in Avian Biology, 14:36–51, 1990. J. F. Piatt, A. M. Harding, M. Shultz, S. G. Speckman, T. I. van Pelt, G. S. Drew, and A. B. Kettle. Seabirds as indicators of marine food supplies: Cairns revisited. Marine Ecology Progress Series, 352:221–234, 2007. S. Piper. Survival of adult, territorial longtailed wagtails Motacilla clara: the effects of environmental factors and individual covariates. Journal of Applied Statistics, 29(1-4):107–124, 2002. K. H. Pollock. Capture-recapture models allowing for age-dependent survival and capture rates. Biometrics, 37:521–529, 1981. K. H. Pollock, J. D. Nichols, C. Brownie, and J. E. Hines. Statistical inference for capture-recapture experiments. Wildlife Monographs, 107:1–97, 1990. E. S. Poloczanska, R. C. Babcock, A. Butler, A. J. Hobday, O. Hoegh-Guldberg, T. J. Kunz, R. Matear, D. Milton, T. A. Okey, A. J. Richardson, et al. Climate change and Australian marine life. Oceanography and Marine Biology, 45:407– 478, 2007. E. S. Poloczanska, C. J. Brown, W. J. Sydeman, W. Kiessling, D. S. Schoeman, P. J. Moore, K. Brander, J. F. Bruno, L. B. Buckley, M. T. Burrows, et al. Global imprint of climate change on marine life. Nature Climate Change, 3(10):919–925, 2013. T. J. Preston, Y. Ropert-Coudert, A. Kato, A. Chiaradia, R. Kirkwood, P. Dann, and R. D. Reina. Foraging behaviour of little penguins Eudyptula minor in an artificially modified environment. Endangered Species Research, 4(1):95–103, 2008. R Development Core Team. R: A Language and Environment for Statistical Com- puting. R Foundation for Statistical Computing, Vienna, Austria, 2010. URL http://www.R-project.org/. ISBN 3-900051-07-0. R. M. Randall, B. M. Randall, J. Cooper, G. D. L. Cock, and G. J. B. Ross. Jack- ass penguin Spheniscus demersus movements, inter-island visits, and settlement (movimientos, visitas inter-islas y establecimiento de residencia en el ping¨uino spheniscus demersus). Journal of Field Ornithology, 58(4):445–455, 1987. N. Ratcliffe, P. Catry, K. C. Hamer, N. I. Klomp, and R. W. Furness. The effect of age and year on the survival of breeding adult great skuas Catharacta skua in Shetland. Ibis, 144(3):384–392, 2002. B. Raymond, S. A. Shaffer, S. Sokolov, E. J. Woehler, D. P. Costa, L. Einoder, M. Hindell, G. Hosie, M. Pinkerton, P. M. Sagar, et al. Shearwater foraging in the Southern Ocean: the roles of prey availability and winds. PloS One, 5(6): e10960, 2010.

205 K. Reid and J. P. Croxall. Environmental response of upper trophic-level predators reveals a system change in an Antarctic marine ecosystem. Proceedings of the Royal Society B, 268(1465):377–384, 2001. K. Reid, J. P. Croxall, D. R. Briggs, and E. J. Murphy. Antarctic ecosystem monitoring: quantifying the response of ecosystem indicators to variability in Antarctic krill. ICES Journal of Marine Science: Journal du Conseil, 62(3): 366–373, 2005. T. Reid, M. Hindell, J. L. Lavers, and C. Wilcox. Re-examining mortality sources and population trends in a declining seabird: using Bayesian methods to incor- porate existing information and new data. PloS One, 8(4):e58230, 2013. P. N. Reilly and J. M. Cullen. The little penguin Eudyptula minor in Victoria. I: mortality of adults. Emu, 79:97–102, 1979. P. N. Reilly and J. M. Cullen. The little penguin Eudyptula minor in Victoria. II: Breeding. Emu, 81:1–19, 1981. doi:10.1071/MU9810001. P. N. Reilly and J. M. Cullen. The little penguin Eudyptula minor in Victoria. III: Dispersal of chicks and survival after banding. Emu, 82:137–142, 1982. P. N. Reilly and J. M. Cullen. The little penguin Eudyptula minor in Victoria. IV: Moult. Emu, 83:94–98, 1983. M. Renner and L. S. Davis. Survival analysis of little penguin Eudyptula minor chicks on Motuara Island, New Zealand. Ibis, 143(4):369–379, 2001. R. W. Reynolds and T. M. Smith. Improved global sea surface temperature analysis using optimum interpolation. Journal of Climate, 7(6):929–948, 1994. K. Ridgway. Long-term trend and decadal variability of the southward penetration of the East Australian Current. Geophysical Research Letters, 34(13), 2007. S. Robinson, A. Chiaradia, and M. A. Hindell. The effect of body condition on the timing and success of breeding in little penguins Eudyptula minor. Ibis, 147(3): 483–489, 2005. Y. Ropert-Coudert, A. Kato, Y. Naito, and B. L. Cannell. Individual diving strate- gies in the little penguin. Waterbirds, 26(4):403–408, 2003. Y. Ropert-Coudert, B. Cannell, and A. Kato. Temperature inside nest boxes of little penguins. Wildlife Society Bulletin, 32(1):177–182, 2004a. Y. Ropert-Coudert, R. P. Wilson, F. Daunt, and A. Kato. Patterns of energy acquisition by a central place forager: benefits of alternating short and long foraging trips. Behavioral Ecology, 15(5):824–830, 2004b. Y. Ropert-Coudert, A. Kato, R. P. Wilson, and B. Cannell. Foraging strategies and prey encounter rate of free-ranging little penguins. Marine Biology, 149(2): 139–148, 2006. Y. Ropert-Coudert, A. Kato, and A. Chiaradia. Impact of small-scale environmental perturbations on local marine food resources: a case study of a predator, the little

206 penguin. Proceedings of the Royal Society of London B, 276:4105–4109, 2009. doi: 10.1098/rspb.2009.1399. Y. Ropert-Coudert, A. Kato, X. Meyer, M. Pell´e,A. J. MacIntosh, F. Angelier, O. Chastel, M. Widmann, B. Arthur, B. Raymond, et al. A complete breeding failure in an Ad´eliepenguin colony correlates with unusual and extreme environ- mental events. Ecography, 38(2):111–113, 2015. R. R. Rykaczewski and D. M. Checkley. Influence of ocean winds on the pelagic ecosystem in upwelling regions. Proceedings of the National Academy of Sciences, 105(6):1965–1970, 2008. B.-E. Sæther, R. Andersen, and H. C. Pedersen. Regulation of parental effort in a long-lived seabird an experimental manipulation of the cost of reproduction in the antarctic petrel, Thalassoica antarctica. Behavioral Ecology and Sociobiology, 33(3):147–150, 1993. P. Sagar, J. Molloy, H. Weimerskirch, and J. Warham. Temporal and age-related changes in survival rates of southern Buller’s albatrosses Thalassarche bulleri bulleri at the Snares, New Zealand, 1948 to 1997. The Auk, 117(3):699–708, 2000. N. Saino, D. Rubolini, N. Jonzn, T. R. N. Ergon, A. Montemaggiori, N. C. Stenseth, and F. Spina. Temperature and rainfall anomalies in Africa predict timing of spring migration in trans-Saharan migratory birds. Climate Research, 35(1):123, 2007. M. Salton, C. Saraux, P. Dann, and A. Chiaradia. Carry-over body mass effect from winter to breeding in a resident seabird, the little penguin. Royal Society Open Science, 2(1):140390, 2015. P. A. Sandery. Seasonal variability of water mass properties in Bass Strait: three- dimensional oceanographic modelling studies. PhD thesis, Flinders University, Adelaide, 2007. P. A. Sandery and J. K¨ampf. Winter-spring flushing of Bass Strait, southeastern Australia: a numerical modelling study. Estuarine, Coastal and Shelf Science, 63 (1-2):23–31, 2005. H. Sandvik, K. E. Erikstad, R. T. Barrett, and N. G. Yoccoz. The effect of climate on adult survival in five species of North Atlantic seabirds. Journal of Animal Ecology, 74:817–831, 2005. H. Sandvik, T. I. M. Coulson, and B.-E. Sæher. A latitudinal gradient in climate effects on seabird demography: results from interspecific analyses. Global Change Biology, 14(4):703–713, 2008.

207 C. Saraux, C. Le Bohec, J. M. Durant, V. A. Viblanc, M. Gauthier-Clerc, D. Beaune, Y.-H. Park, N. G. Yoccoz, N. C. Stenseth, and Y. Le Maho. Re- liability of flipper-banded penguins as indicators of climate change. Nature, 469: 203–206, 2011. C. Saraux, A. Chiaradia, M. Salton, P. Dann, and V. A. Viblanc. Negative effects of wind speed on individual foraging performance and breeding success in little penguins. Ecological Monographs, 86(1):61–77, 2016. H.-C. Schaefer, G. W. Eshiamwata, F. B. Munyekenye, E. Griebeler, and K. B¨ohning-Gaese. Monthly survival of African Sylvia warblers in a seasonally arid tropical environment. Ibis, 148(3):411–424, 2006. R. Schahinger. Structure of coastal upwelling events observed off the southeast coast of South Australia during February 1983–April 1984. Marine and Freshwater Research, 38(4):439–459, 1987. I. D. Schroeder, W. J. Sydeman, N. Sarkar, S. A. Thompson, S. J. Bograd, and F. B. Schwing. Winter pre-conditioning of seabird phenology in the California Current. Marine Ecology Progress Series, 393:211–223, 2009. N. Schumann, P. Dann, and J. P. Y. Arnould. The significance of northern-central Bass Strait in southeastern Australia as habitat for burrowing seabirds. Emu, 114(3):234–240, 2014. C. J. Schwarz and A. N. Aranson. A general methodology for the analysis of capture-recapture experiments in open populations. Biometrics, 52(3):860–873, 1996. G. A. F. Seber. A note on the multiple recapture census. Biometrika, 52:249–259, 1965. doi:10.1093/biomet/52.1-2.249. G. A. F. Seber. Estimating time-specific survival and reporting rates for adult birds from band returns. Biometrika, 57:313–318, 1970. G. A. F. Seber. Estimating age-specific survival rates from bird-band returns when the reporting rate is constant. Biometrika, 58:491–497, 1971. G. A. F. Seber and C. J. Schwarz. Capture-recapture: Before and after EURING 2000. Journal of Applied Statistics, 29(1-4):5–18, 2002. C. H. Sekercio˘glu,R. B. Primack, and J. Wormworth. The effects of climate change on tropical birds. Biological Conservation, 148(1):1–18, 2012. R. B. Sherley, B. J. Barham, P. J. Barham, T. M. Leshoro, and L. G. Underhill. Artificial nests enhance the breeding productivity of African penguins Spheniscus demersus on Robben Island, South Africa. Emu, 112(2):97–106, 2012a. R. B. Sherley, K. Ludynia, L. G. Underhill, R. Jones, and J. Kemper. Storms and heat limit the nest success of bank cormorants: implications of future climate change for a surface-nesting seabird in southern Africa. Journal of Ornithology, 153(2):441–455, 2012b.

208 L. A. Sidhu. Analysis of recovery-recapture data for little penguins. PhD thesis, The University of New South Wales, Canberra, 2007. L. A. Sidhu, E. A. Catchpole, and P. Dann. Mark-recapture-recovery modeling and age-related survival in little penguins Eudyptula minor. The Auk, 124(3): 815–827, 2007. L. A. Sidhu, P. Dann, L. Chambers, and E. Catchpole. Seasonal ocean temperature and the survival of first-year little penguins Eudyptula minor in southeastern Australia. Marine Ecology Progress Series, 454:265–272, 2012. J. R. Skalski. Regression of abundance estimates from mark recapture surveys against environmental covariates. Canadian Journal of Fisheries and Aquatic Sciences, 53(1):196–204, 1996. J. Sommerfeld, A. Kato, Y. Ropert-Coudert, S. Garthe, C. Wilcox, and M. A. Hindell. Flexible foraging behaviour in a marine predator, the masked booby Sula dactylatra, according to foraging locations and environmental conditions. Journal of Experimental Marine Biology and Ecology, 463(0):79–86, 2015. L. B. Spear and D. G. Ainley. Flight speed of seabirds in relation to wind speed and direction. Ibis, 139(2):234–251, 1997. E. B. Spurr. Behavior of the Ad´eliepenguin chick. The Condor, 77(3):272–280, 1975. C. Stahel and R. Gales. Little Penguins: Fairy Penguins in Australia. New South Wales University Press, Kensington, 1987. C. Stahel and S. C. Nicol. Ventilation and oxygen extraction in the little pen- guin Eudyptula minor, at different temperatures in air and water. Respiration Physiology, 71(3):387–398, 1988. C. D. Stahel and S. C. Nicol. Temperature regulation in the little penguin, Eudyptula minor, in air and water. Journal of Comparative Physiology B, 148:93–100, 1982. S. C. Stearns. Life-history tactics: a review of the ideas. Quarterly review of biology, pages 3–47, 1976. S. C. Stearns. The Evolution of Life Histories. Oxford University Press, Oxford, 1992. C. Stevenson and E. J. Woehler. Population decreases in little penguins Eudyptula minor in southeastern Tasmania, Australia, over the past 45 years. Marine Ornithology, 35(1):71–76, 2007. A. Stier, V. A. Viblanc, S. Massemin-Challet, Y. Handrich, S. Zahn, E. R. Rojas, C. Saraux, M. Le Vaillant, O. Prud’Homme, E. Grosbellet, et al. Starting with a handicap: phenotypic differences between early-and late-born king penguin chicks and their survival correlates. Functional ecology, 28(3):601–611, 2014. D. L. Stokes and P. D. Boersma. Nest-site characteristics and reproductive success in Magellanic penguins Spheniscus magellanicus. The Auk, 115(1):34–49, 1998.

209 R. M. Suryan, D. J. Anderson, S. A. Shaffer, D. D. Roby, Y. Tremblay, D. P. Costa, P. R. Sievert, F. Sato, K. Ozaki, G. R. Balogh, et al. Wind, waves, and wing loading: morphological specialization may limit range expansion of endangered albatrosses. PLoS One, 3(12):e4016, 2008. D. R. Sutherland and P. Dann. Population trends in a substantial colony of lit- tle penguins: three independent measures over three decades. Biodiversity and Conservation, 23(1):241–250, 2014. W. J. Sydeman, M. M. Hester, J. A. Thayer, F. Gress, P. Martin, and J. Buffa. Climate change, reproductive performance and diet composition of marine birds in the southern California Current system, 1969-1997. Progress in Oceanography, 49(1):309–329, 2001. W. J. Sydeman, S. A. Thompson, and A. Kitaysky. Seabirds and climate change: roadmap for the future. Marine Ecology Progress Series, 454:107–117, 2012. K. L. Szostek and P. H. Becker. Survival and local recruitment are driven by environmental carry-over effects from the wintering area in a migratory seabird. Oecologia, 178(3):643–657, 2015. G. Tavecchia, R. Pradel, F. Gossmann, C. Bastat, Y. Ferrand, and J.-D. Lebre- ton. Temporal variation in annual survival probability of the Eurasian woodcock Scolopax rusticola wintering in France. Wildlife Biology, 8(1):21–30, 2002. J.-B. Thiebot, Y. Cherel, M. Acqueberge, A. Prudor, P. N. Trathan, and C.-A. Bost. Adjustment of pre-moult foraging strategies in Macaroni penguins Eudyptes chrysolophus according to locality, sex and breeding status. Ibis, 156(3):511–522, 2014. M. K. Thomas, C. T. Kremer, C. A. Klausmeier, and E. Litchman. A global pattern of thermal adaptation in marine phytoplankton. Science, 338(6110):1085–1088, 2012. P. M. Thompson and V. Grosbois. Effects of climate variation on seabird population dynamics. Directions in Science, 1:50–52, 2002. H. V. Thurman and A. P. Trujillo. Introductory Oceanography. Pearson Education Inc, Upper Saddle River, 10th edition, 2004. T. Tveraa, B.-E. Sether, R. Aanes, and K. E. Erikstad. Regulation of food pro- visioning in the antarctic petrel; the importance of parental body condition and chick body mass. Journal of Animal Ecology, 67(5):699–704, 1998. S. Verhulst and J.-A.˚ Nilsson. The timing of birds’ breeding seasons: a review of experiments that manipulated timing of breeding. Philosophical Transactions of the Royal Society of London B: Biological Sciences, 363(1490):399–410, 2008. C. M. Vleck and D. Vleck. Physiological condition and reproductive consequences in ad´eliepenguins. Integrative and Comparative Biology, 42(1):76–83, 2002.

210 A. Von Br¨omssenand C. Jansson. Effects of food addition to willow tit Parus montanus and crested tit P. cristatus at the time of breeding. Ornis Scandinavica, 11(3):173–178, 1980. A. Wald. Tests of statistical hypotheses concerning several parameters when the number of observations is large. Transactions of the American Mathematical Society, 54:426–482, 1943. T. M. Ward, L. J. McLeay, W. F. Dimmlich, P. J. Rogers, S. A. M. McClatchie, R. Matthews, J. K¨ampf,and P. D. Van Ruth. Pelagic ecology of a northern boundary current system: effects of upwelling on the production and distribution of sardine Sardinops sagax, anchovy Engraulis australis and southern bluefin tuna Thunnus maccoyii in the Great Australian Bight. Fisheries Oceanography, 15(3): 191–207, 2006. B. W. Weavers. Seasonal foraging ranges and travels at sea of little penguins Eu- dyptula minor, determined by radiotracking. Emu, 91(5):302–317, 1992. M. S. Weerheim, N. I. Klomp, A. M. Brunsting, and J. Komdeur. Population size, breeding habitat and nest site distribution of little penguins Eudyptula minor on Montague Island, New South Wales. Wildlife Research, 30(2):151–157, 2003. H. Weimerskirch. Reproductive effort in long-lived birds: age-specific patterns of condition, reproduction and survival in the wandering albatross. Oikos, 64(3): 464–473, 1992. H. Weimerskirch, J. C. Stahl, and P. Jouventin. The breeding biology and popula- tion dynamics of King Penguins Aptenodytes patagonicus on the Crozet Islands. Ibis, 134:107–117, 1992. H. Weimerskirch, T. Guionnet, J. Martin, S. A. Shaffer, and D. Costa. Fast and fuel efficient? Optimal use of wind by flying albatrosses. Proceedings of the Royal Society of London B: Biological Sciences, 267(1455):1869–1874, 2000. H. Weimerskirch, L. Zimmermann, and P. A. Prince. Influence of environmental variability on breeding effort in a long-lived seabird, the yellow-nosed albatross. Behavioral Ecology, 12(1):22–30, 2001. H. Weimerskirch, P. Inchausti, C. Guinet, and C. Barbraud. Trends in bird and seal populations as indicators of a system shift in the Southern Ocean. Antarctic Science, 15(2):249–256, 2003. H. Weimerskirch, M. Louzao, S. de Grissac, and K. Delord. Changes in wind pattern alter albatross distribution and life-history traits. Science, 335(6065): 211–214, 2012. G. C. White and K. P. Burnham. Discussion comments on: the use of auxiliary vari- ables in capture-recapture modelling: an overview. Journal of Applied Statistics, 29:103–106, 2002.

211 R. J. Whittington, J. B. Jones, P. M. Hine, and A. D. Hyatt. Epizootic mortality in the pilchard Sardinops sagax neopilchardus in Australia and New Zealand in 1995. i. pathology and epizootiology. Diseases of Aquatic Organisms, 28(1):1–15, 1997. B. Wienecke, J. Bradley, and R. Wooller. Annual and seasonal variation in the growth rates of young little penguins Eudyptula minor in Western Australia. Emu, 100(2):139–147, 2000. A. J. Williams, W. R. Siegfried, A. E. Burger, and A. Berruti. Body composition and energy metabolism of moulting Eudyptid penguins. Comparative Biochemistry and Physiology Part A: Physiology, 56(1):27–30, 1977. K. Williamson. Birds and climatic change. Bird Study, 22(3):143–164, 1975. R. P. Wilson. Animal behaviour: The price tag. Nature, 469(7329):164–165, 2011. G. Yannic, A. Aebischer, B. Sabard, and O. Gilg. Complete breeding failures in ivory gull following unusual rainy storms in North Greenland. Polar Research, 33:22749, 2014. URL http://dx.doi.org/10.3402/polar.v33.22749. A. J. Zera and L. G. Harshman. The physiology of life history trade-offs in animals. Annual Review of Ecology and Systematics, 32:95–126, 2001.

212