Sensor-based parturition detection in ewes: Towards autonomous welfare assessment for sheep

The livestock industry is facing one of its biggest challenges to date: to increase productivity and meet the world’s growing demand for protein, yet at the same time improve sustainability outcomes, particularly those related to social license and animal welfare. This is not a simple challenge to overcome, with the demands of increased production and animal welfare often seen as competing. However, there is one suite of technologies which may hold the answer to this problem: on-animal sensors. Many of these systems are in their commercial infancy and research is required to provide guidance on how they may be applied to solve critical industry problems linked to production and welfare. A key component of this research is exploring how the raw outputs from these sensors can be converted from large complex data to meaningful information so that a producer is alerted to a problem and can implement an intervention strategy. This thesis examines the potential for on-animal sensor technology for autonomous detection of a key event in grazing sheep production systems: parturition. Parturition (or lambing) can be considered one of the most important periods in the breeding animal’s life. It is a period of vulnerability for the mother and newborn, requiring specific physical, physiological and behavioural changes to ensure survival. Parturition is also a period of significant welfare risk both to the ewe (in the form of disease, particularly dystocia) and the lamb (from mismothering and a range of other issues). Chapter 1 is a general introduction and briefly introduces the major concepts that will be addressed. Chapter 2 is a published manuscript and provides a general review of the use of sensor technology in sheep research. It provides an understanding of which sensors have been applied to sheep and the likely best candidates for use in my research. Chapter 3 extends this knowledge and explores how sensor technologies might be employed in welfare assessment. This chapter has also been published. Through this scoping work (Chapters 2 and 3), Global Navigation Satellite System (GNSS) tracking and accelerometers were identified as two key sensor systems. Both are readily available and likely to provide the required information for detecting parturition-related behaviours. Since the experimental chapters (Chapters 5 - 9) were developed with the intention of publication, it was not possible to provide in-depth information on the complete data analysis process. Consequently, an additional chapter (Chapter 4) was included prior to the experimental chapters to provide more background and context for the thesis reader. This chapter outlines the data analysis methods used throughout the thesis. Chapters 5, 6 and 7 explore how data from GNSS and accelerometers can be interpreted and related to the behavioural changes expressed by sheep around parturition. Chapter 5 is a published manuscript and reports on the value of GNSS tracking for the detection of lambing-related behaviours. This research demonstrated the value of GNSS data to monitor daily changes in movement, social activity and landscape utilisation associated with parturition. However, at the time of publication, limitations in the ability of GNSS to detect hourly changes in behaviour were evident, and so the data from accelerometer sensors were also explored. Contrasting with the results of this chapter, later analysis uncovered further potential of GNSS to detect hourly changes in behaviour using novel metrics. To understand the potential for accelerometers to detect parturition activity, basic behavioural algorithms were first developed using machine learning (ML) classification techniques (Chapter 6). These results have been published and indicate the ability of ML to detect common behaviours from accelerometer data with an accuracy of 76.9 – 98.1 %. These algorithms were then applied in Chapter 7 to explore if the accelerometer data could be linked to parturition-related activities. Similar to Chapter 5, measurable changes in behaviour were identified and associated with parturition. In particular, ewes significantly increased their walking behaviour and the number of posture changes in the hour of parturition. This chapter has also been published. The ultimate goal of this thesis was to determine if a near-real-time sensor-based system might be developed to detect parturition in ewes, specifically via remote monitoring of typical parturition behaviours. Chapter 8 integrates the knowledge gained in Chapters 5 and 7, building a simulated near-real-time parturition detection model that integrates the two sensor types. Overall, the final model successfully identified between 81.8% and 90.9 % of lambing events within ± 3 h of known birth, with accuracy depending on the use of different alert criteria. Chapter 9 has been prepared as a short communication. This chapter applies the model developed in Chapter 8 and investigates the value of the model for the assessment of welfare at lambing using data gathered from an adverse lambing event (prolapsed animal). The results suggest that ewes with repeated lambing alerts that are not followed by parturition may be at-risk of an adverse event and should be closely inspected by the producer. The final chapter (Chapter 10) is the synthesis chapter, reporting research conclusions, limitations and recommendations for future research.