Toby Patterson

Statistical methods for the analysis of electronic tag data: From individuals to populations

Supervisors: Dr Mark Bravington (CMAR) and A/Prof Mark Hindell (UTAS)

Background

The use of electronic tags (archival tags and satellite tags) to study the movements and behaviour of a variety of pelagic species has increased rapidly in recent years. Tag data consists of a multivariate time series data of which, due to the complexity, volume and detail of the data, present a challenge for investigators. Statistically robust methods are lacking to answer such basic questions as:

• How can we objectively classify and predict behaviour from electronic tag datasets?
• Do different individuals behave in a significantly different way with respect to environmental covariates?
• Can we make general inferences about and better predict the horizontal and vertical distribution of animals using electronic tag data?

At present there is no accepted statistical methodology to handle such fundamental questions and the traditional statistical paradigm for analysis of ecological data is of little use in addressing these specific problems due to the highly autocorrelated and nonlinear data. Furthermore, each of these fundamental questions is crucial in larger scale questions about the role of top-predators in the pelagic ecosystem including:

• How do animals move and search for prey in a pelagic environment?
• How would large scale changes to oceanography change distribution and habitat usage of pelagic species?
• Determining decision rules for Individual Based Movement Models
• Building robust habitat preference models for management applications
• Can spatial management effectively manage “hotspots” of pelagic predator abundance?
• Testing the predictions of foraging theory

Each of these questions requires quantification of behaviour and habitat preference and characterization of the extrinsic factors driving them.

Objectives

The project will examine the use of new statistical methodology such as Hidden Markov Models (commonly used in artificial intelligence, speech recognition and genomics research) and Bayesian Hierarchical Models to develop a new approach to the analysis of marine telemetry data. The project will develop a meta-analysis of marine predator behaviour in relation to environmental datasets and investigate robust ways to scale up from individuals to populations.