PDF (Whole thesis) sumner.pdf | Download (3MB) Available under University of Tasmania Standard License.

Abstract

This thesis provides an integrated approach to common problems with location
estimation for animal tracking. Many existing techniques confound problems
of location accuracy with simplistic track representations and under-utilization of
available data.
Traditional techniques such as speed ltering and time spent maps are illustrated
with a software package developed by the author with examples of location
estimates from southern elephant seals. This software enables the application and
exploration of various techniques that have previously not been available in a single
solution. These include ltering, temporal gridding, projection transformation and
GIS integration.
A novel Bayesian approach is introduced for the more general problems faced
by dierent tagging techniques. This approach integrates all sources of data including
movement models, environmental data and prior knowledge. This general
framework is illustrated by application to satellite tag data and light-measuring tag
data. Examples are used to detail the use of movement models with a powerful track
representation model, and the application of raw light data for location estimation.
Previously under-utilized sources of data are used to inform location estimates.
A method for applying light level geo-location within the framework is presented.
This approach provides a primary location estimate for each twilight and utilizes all
of the available data from archival tags.
These model runs result in very large databases of samples from Markov Chain
Monte Carlo (MCMC) simulations and techniques for summarizing these for the
variety of analysis outputs are illustrated. This system solves issues of location
uncertainty with a full path representation and provides spatial maps of residency
for multiple animals.
The relation between archival tag data and ocean circulation is used to extend
the application of archival tag data for location estimation for diving animals in a
manner similar to commonly used SST methods. Diving proles from elephants seals
are compared with 4D oceanographic datasets. Older tags are limited by problems
with measurement lags for temperature|this problem is addressed with a proxy
model for temperature at depth to ocean height.
This thesis provides a number of important improvements to the derivation of
location from various types of tag data by integrating disparate information sources
in a systematic way. Location estimates are produced with inherent quantication
of errors. The approach provides the variety of metrics and analysis types required
with an extensible software package. These contributions help bridge the divides
between various analytic techniques traditionally employed for animal tracking.

Item Type:	Thesis - PhD
Authors/Creators:	Sumner, MD
Keywords:	animal tracking, Bayesian, MCMC, light level geo-location, Argos
Copyright Holders:	The Author
Copyright Information:	Copyright 2011 the Author.
Item Statistics:	View statistics for this item

Actions (login required)

Item Control Page