| Summary: | The impact of lameness on animal welfare and productivity is a significant concern in dairy farming. Claw Horn Disruption Lesions (CHDLs) are the leading cause of lameness however significant gaps exist in our current understanding of their temporal risk relation, detection, and behavioural implications. The availability of long-term, large-scale datasets combined with the effective use of emerging technologies, particularly sensors for behavioural classification, enables a more confident expansion on the current literature and a deeper exploration.
The primary objective of this thesis is to harness the power of big data and modern advanced data modelling techniques to address the current challenges and provide insight into prevention and detection of CHDLs.
Chapter 3 of this thesis employs a comprehensive longitudinal model of lameness risks to temporally relate the first lesion event to the longer-term trajectory of a cow’s claw health. It demonstrates that preventing the first lesion from occurring within 180 days of the first calving event results in long-term benefits for a cow’s claw health. Additionally, it shows that the type and severity of the first lesion event affects the frequency and period of later lesion events. The potential of Mid Infra-Red (MIR) milk spectroscopy data for detecting CHDLs in dairy cattle is then evaluated and illustrates how the use of Generative Adversarial Networks (GANs) can be used to mitigate the effect of confounders within large datasets. Chapter 5 describes the quantification of within and between cow behavioural variability for lying and eating behaviours in adult dairy cattle for the first time. As a potential source of the lack of consensus among studies focussed on the effects of CHDLs and lameness on behaviour, this work highlights the influence of individual cow personality on behavioural data. Finally, the thesis employs dynamic temporal features of behavioural data to quantify the impact of CHDLs on the periodicity lying and eating behaviours and reveals inherent differences in cow behaviour between those that will eventually be exposed to CHDLs and those that are not.
Overall, this thesis uses demonstrates the usefulness of modern large scale data repositories and how novel results can be extracted from data already available on thousands of commercial dairy farms across the globe.
|
|---|