Development and validation of animal-based welfare indicators for a Precision Livestock Farming (PLF) approach to small ruminant welfare management

Abstract

To improve animal welfare management on sheep farms, it must first be measured through reliable welfare indicators. In extensive management systems, gathering welfare information is labour intensive and time-consuming. In a rural environment where the labour force is dwindling and the average age of farmers is climbing, technology is often declared to be an all-encompassing solution. Precision livestock farming (PLF) is defined as managing individual animals by continuous real-time monitoring of health, welfare, production, reproduction, and environmental impact, essentially every aspect of a farmed animal’s life. It utilises all types of technology to allow farmers to collect information efficiently and accurately about individual animals. In sheep farming, commercialised PLF tools are rare. However, sheep behaviour has been reported to change when welfare challenges occur. For example, grazing patterns are altered by gastrointestinal parasitism and lameness affects the smoothness of movements. Therein lies an opportunity to use technology to monitor sheep behaviour and identify when welfare challenges occur in extensively raised sheep. However, PLF development and roll-out face challenges such as high costs, rare validation and the ethical questions surrounding increased automation in animal farming.

This thesis is composed of a series of experiments on domestic sheep (Ovis aries) to test the accuracy of potential welfare indicators and the ability of two PLF tools to record them. I hypothesised that accelerometers and Bluetooth beacons would be able to measure behavioural changes in sheep experiencing lameness, gastrointestinal parasitism and mastitis, as these are some of the most important welfare challenges faced by extensively managed sheep. A pilot trial was undertaken indoors to examine if any behavioural changes occurred during gastrointestinal parasitism that could eventually be monitored by PLF approaches (Chapter 2). This experiment took advantage of a large parasitological trial, organised for other purposes, to look for behavioural differences between lambs parasitised with Teladorsagia circumcincta and healthy lambs. Behaviour was monitored through video recordings that were scan sampled and behaviour sampled, and in-person using Qualitative Behaviour Assessment (QBA). Lambs were separated into three treatment groups: 1) ad-lib fed controls (AC), 2) restricted-fed controls (RC) and 3) ad-lib fed parasitised lambs (AP). Parasitised lambs were found to be more likely to be standing inactive than AC lambs and less likely to be eating than RC lambs over the first 3 weeks of infection. Parasitised lambs had higher loadings on the QBA dimension describing fear and anxiety compared to RC lambs. These results were interpreted as reflecting the discomfort caused by abomasal damage inflicted by T.circumcincta larvae and the expected parasite-induced anorexia. Chapter 2 also offered novel evidence that not only did parasitism negatively impact lamb health but it also affected their mental state by increasing levels of fear and anxiety.

Chapter 3 describes trials occurring over two grazing seasons where 56 ewes and 112 lambs faced natural infection with lameness, mastitis, and gastrointestinal parasites. Daily scan sampling, weekly QBA and fortnightly welfare assessments monitored their behaviour and welfare. Welfare assessments consisted of recording weights, dag scores, fleece, breathing and injury scores for all sheep, and additional body condition scoring and dentition scores for ewes. Lameness scoring, faecal egg counts (FEC), and mastitis scoring by udder palpation for ewes monitored the diseases of interest during both years, with somatic cell counts (SCC) of ewe milk samples being added in the second year of the trial. Generalised linear mixed models (GLMM) were used to analyse the relationships between behaviour and welfare indicators. Grazing behaviour in lambs was significantly associated with lamb lameness and parasitism. There was a significant interaction between lamb lameness score and strongyle FEC that affected their locomotion and lying behaviour. Nematodirus FEC had a significant impact on lamb lying, standing and play. Ewe lameness was associated with lying behaviour. Chapter 3 concluded that ewe and lamb parasitism and lameness had the potential to be identified through behavioural indicators.

Chapter 4 aimed to validate the use of AX3 accelerometers (Axivity Ltd., Newcastleupon-Tyne, UK) to categorise ewe and lamb behaviour. To do this, 6 ewes and 6 lambs were observed using 20-minute focal samples on 4 days. Their recorded behaviours were compared to the AX3 outputs. This chapter also tested whether wearing a collar containing the AX3 had any impact on sheep behaviour or welfare. This was done using data from the trials in Chapter 3 since half of the animals were wearing collars containing technology while the other half acted as a control group, not wearing collars. Results showed that ewe rumination was less likely to be observed when they were wearing collars. Ewes and lambs had a higher probability of presence of strongyle eggs in faecal samples when they were not wearing collars. This implies that grazing behaviour may have differed between sheep with and without collars, leading to increased exposure to strongyle larvae for the “no collar” control group. The AX3 validation was attempted using a series of statistical methods. Very low levels of variation in the accelerometer data made comparison of behaviours challenging. However, k-means clustering partially categorised some behaviours, such as grazing and standing. The validation was not entirely successful, but it led to the conclusion that unlabelled machine learning techniques may be able to help complete this validation with more variable data and purpose-built algorithms. Further work is required to clarify the findings in this chapter suggesting collars could impact rumination and grazing behaviour.

A second PLF tool was tested in Chapter 5: Bluetooth Light Energy (BLE) beacons (Feasycom, Shenzhen, China). This study tested their ability to monitor ewe-lamb distance as an indicator of welfare. Collars containing the BLE beacons were put on lambs as they were born, while the ewes wore purpose-built readers called Wearable Integrated Sensor Platform (WISP) readers on collars. The WISP readers transmitted a Received Signal Strength Indicator (RSSI) via a low-power wide-area network (LPWAN) gateway every 5 minutes for the 16 beacons closest to it. RSSI was converted into distance in metres and the data was filtered so that only the observations concerning the distance between each ewe and her offspring would remain. Weekly welfare assessments were carried out in person on all animals for 6 weeks after the start of lambing. These assigned binary lameness, fleece and dag scores to every animal (0-absence of welfare issue/1-presence of welfare issue). All lambs had lameness, fleece and dag scores of 0 for the duration of the study, rendering it impossible to draw any conclusions about ewe-lamb distance as an indicator of lamb welfare. However, ewes with a lameness score or a fleece score of 1 had shorter ewelamb distances, meaning their lambs remained closer to them than lambs who had dams with lameness or fleece scores of 0. Chapter 5 concluded that ewe-lamb distance is associated with ewe welfare indicators and can be measured by PLF.

Chapter 6 describes the findings from explorative semi-structured interviews with Norwegian sheep farmers about their use of PLF technology. Norway has a high rate of PLF adoption in livestock farming compared to other European countries, and therefore offered the opportunity to study the motivations and perspectives of producers currently using technology. Twenty-four farmers from three regions in Norway who use a form of technology on their sheep farms were interviewed. The interview was designed to understand what drove their initial adoption of PLF, why they continue to use it or not, and their vision of the future with PLF in sheep farming. Reflexive thematic analysis identified five main themes from the farmers’ responses: Resources and Savings, Control and Decision-making, Governmental Influences and Pressures, Out with the Old and In with the New, and Curiosity and Excitement. Many decisions were driven by farm economics, where PLF improved a costly process or saved the farmers time. Several farmers referred to the increased amount of control that PLF offered them over their flock. The government was seen as a source of both support and hindrance to farmers using PLF. Participants expressed a perceived incompatibility between PLF and older users, although this was not reflected in their reality as many older farmers interviewed for this study invested heavily in technology. And finally, many farmers simply found the extra information they gained was fun, interesting, and satisfying. The farmer motivations identified in this chapter have implications for our understanding of how and why PLF is applied on farm and could inform the development of future technologies.

These findings suggest that behaviours could be used to monitor the welfare of sheep in extensive management systems. Because in-person monitoring is time-consuming, PLF tools have been found to have potential to monitor the changes in behaviour. However, this thesis also reported the importance and challenges of validating such technology. This highlights the need for robust, independent validation studies to support the growing interest in PLF for livestock. The findings related to behavioural indicators and farmer motivations around PLF have implications for the future development of PLF tools for welfare monitoring.