Methods to Improve Our Understanding of the Health and Welfare Status of Sheep (Ovis Aries) and the Influences of their Immediate Environment
Bradley, D
Date: 3 April 2023
Thesis or dissertation
Publisher
University of Exeter
Degree Title
Master of Philosophy in Psychology
Abstract
Studies into the effective use of accelerometers in the automated assessment of sheep behaviour to improve welfare has increased exponentially with promising preliminary results. Previous research has focused primarily on explicit behaviour classification, for example, parturition and urination events, with a view to create a commercial ...
Studies into the effective use of accelerometers in the automated assessment of sheep behaviour to improve welfare has increased exponentially with promising preliminary results. Previous research has focused primarily on explicit behaviour classification, for example, parturition and urination events, with a view to create a commercial tool that will provide health warnings for farmers. Yet the majority of trials have not been conducted in a farm environment. This study aims to provide essential primary research investigating environmental variables that may influence the behavioural patterns of a commercial flock. This vital information has been largely overlooked and crucial when considering tools that provide health warnings, due to the many factors that influence sheep behaviour such as weather, vegetation, soil type, land typography and breed (Hinch, 2017).
The primary aim of this study was to assess the most appropriate model to predict the behaviours of commercial ewes. This was achieved by deploying accelerometers on a commercial flock and simultaneously collecting manual observations and video recordings of flock’s individual activity. The raw acceleration data was processed to create 6 variables. Behaviour classification was also evaluated using three ethograms, each with two mutually exclusive behavioural/postural states: 1. Head Position (head up/down), 2. Posture (standing/lying), 3. Activity (resting/grazing). Three Window setting (3, 5 and 7 seconds) and five machine learning algorithms
4
(Linear Discriminate Analysis (LDA), Classification and Regression Trees (CART), K Nearest Neighbour (KNN), Support Vector Machines (SVM) and Random Forest (RF)) were evaluated. Results indicated a RF with a 7 second window the optimal model across all ethograms. (Accuracy by ethogram; 1) 91.5%, 2) 91.0% and 3) 99.3%).
The secondary aim of this study was to use a Linear Mixed Model (LMM) to investigate the influence of temperature and rainfall on grazing and resting behaviours. This was accomplished by using the initially developed model (RF) on data collected from an unsupervised commercial flock, recorded in a second trial. Results indicated that there was a significant positive relationship between grazing durations and rainfall (p.001), this finding conflicts with previous research observations and is yet unpublished. In addition, prior sheep behaviour research has suggested ‘foraging’ as the dominant activity, results from this trial indicate the dominant daily activity was resting (67% of daily activity).
In conclusion this study highlights the difficultly of defining what ‘normal’ sheep behaviour is and that it is not viable to implement a ‘one-size fits all’ approach. Further research is required in the behavioural assessment for this particularly malleable species.
MPhil Dissertations
Doctoral College
Item views 0
Full item downloads 0