When Sensors Go Wrong: Simulating Errors in PLF Data for Cattle Social Networks and Home Range

Background

Social behaviour is a fundamental aspect of cattle welfare and productivity. In our research group, we have shown that dairy cows form complex and stable social relationships, with friendships that can be traced back to early life. For example, calves raised together tend to spend significantly more time with each other later in life compared with the rest of the herd. Furthermore, kinship appears to play a role, with related animals showing stronger social bonds than unrelated individuals (see Marina et al., 2024)

To investigate these patterns more deeply and across herds, positioning data from multiple farms is often required. However, even when the same real-time positioning system is installed, accuracy can vary considerably depending on barn layout and technical conditions. Combining data from different barns, therefore, raises the question of how much error can be tolerated before social network or home range analyses become unreliable. This project addresses this gap by explicitly testing how positioning errors influence behavioural inferences in cattle.

Aim

The aim of this project is to determine the minimum accuracy required from real-time positioning systems to reliably investigate social behaviour and home range in dairy cattle. This will be achieved by simulating different types and magnitudes of errors in positioning data generated by ultra-wideband (UWB) technology and assessing their impact on social network analyses and home range estimations. By establishing tolerance limits for error, the study will provide guidelines for the robust use of positioning data in animal behaviour research.

Project description

This is a data-driven project involving:

• Cleaning and analysis of positioning data from sensor-equipped cows
• Simulation of different types and levels of positioning errors
• Construction and analysis of social networks
• Statistical modelling of network structures and home ranges

Learning Outcomes

1. Practical skills in R programming, including data cleaning, visualization, and statistical modelling
2. Training in network analysis, error simulation, and behavioural data interpretation
3. Experience linking precision livestock farming (PLF) technologies with animal behaviour research
4. Insights into the robustness of behavioural metrics under varying data quality conditions

Specifications

• Experience with R will be beneficial, but not mandatory. A custom-developed R package for data manipulation and analysis will be provided, along with full supervision and guidance.
• Suitable for students in animal science, veterinary medicine, or related fields with an interest in animal behaviour and smart farming technologies.