E-Thesis 138 views 82 downloads
A new approach to identifying behaviour in animals using tiger sharks, Galeocerdo cuvier, as a test species / WALLIS JAMES
Swansea University Author: WALLIS JAMES
PDF | E-Thesis – open access
Copyright: The author, Wallis M. James, 2022.Download (3.76MB)
There are many advantages in determining animal behaviour for conservation initiatives seeking to protect species impacted by the changing planet. However, for many years, direct observation of elusive or dangerous animals in challenging habitats precluded the acquisition of representative, non-bias...
|Degree level:||Master of Research|
|Supervisor:||Wilson, Rory P.|
No Tags, Be the first to tag this record!
There are many advantages in determining animal behaviour for conservation initiatives seeking to protect species impacted by the changing planet. However, for many years, direct observation of elusive or dangerous animals in challenging habitats precluded the acquisition of representative, non-biased behavioural data. Recently though, animal-attached tag technology incorporating accelerometers, magnetometers and pressure sensors, has greatly advanced our abilities to document the behaviour of a numerous vertebrate species (e.g. birds, reptiles, fish, mammals), even when they cannot be observed. For this, supervised and unsupervised machine learning are often used to categorise behaviours by identifying patterns within the biotelemetry data. However, supervised machine learning requires training data, which is not always available, and both methods are driven by machine-based software with no explicitly defined parameters associated with behaviours which can be cross-checked. This work aimed to use a proper physics-based understanding of triaxial accelerometer-, magnetometer- and pressure data taken from electronic tags deployed on tiger sharks, Galeocerdo cuvier, to interpret patterns and group them into behaviours. As part of this, multi-modality in frequency distributions of parameters was investigated, on the premise that different behaviours can result in different frequency distributions in particular metrics. Following examination, algorithms using defined numerical limits were created to isolate distinct behaviours and these used to detect the extent of identified patterns within entire data sets and across individuals. A total of 12,338 minutes of tag data was processed, from which 10 behaviours were identified. Seven of these were successfully described using numerical metric limits from recorded and/or derived data including; ‘descent’, ‘ascent’, ‘burst power’, etc. However, frequency distributions showed a continuum rather than multiple distinct modes, indicating that this approach is likely to be more complex than thought. The use of physical principles seems a promising method for interpreting accelerometer, magnetometer and pressure data to identify behaviours that occur in study animals that cannot be directly observed. Although these algorithms are specific to tiger sharks in this work, this method is likely to be applicable to other species in aerial, aquatic or terrestrial habitats and could inform a broad range of conservations initiatives in the future.
ORCiD identifier: https://orcid.org/0000-0002-8541-3976
Tiger shark, accelerometer tag, behaviour, quantitative, acceleration data, VeDBA
Faculty of Science and Engineering