Sheep can be exposed to a variety of challenges and failure to adapt to these challenges can compromise their health and wellbeing. Regular monitoring of stock on large-scale or extensive systems may not always be possible, although recent technological advancements in automated data capture, such as walk-over-weighing (WoW), can make monitoring easier. The potential benefit of including behavioural assessment in such a system has yet to be tested. We investigated whether quantitative and qualitative behavioural assessment (QBA) methods could be applied to short video footage collected automatically within a WoW setup, to differentiate between sheep that were, presumably, in different (positive and negative) welfare states. Video footage was collected from 36 Merino sheep within the four treatment groups; Control (n = 12), Habituated to the WoW setup and human interaction (n = 8), Lame (n = 8) and Inappetent (n = 8). Habituated sheep were exposed to a low-stress handling regimen for six consecutive days prior to filming. At the same time, feeding behaviour was recorded by means of radio-frequency identification (RFID) technology to identify sheep suffering inappetence. Lame sheep were identified using a 6-point scoring system, and Control animals were selected ensuring that they were not Lame, Inappetent or Habituated. For QBA, the footage was presented, in a random order, to 18 observers. There was a significant (P < 0.001) consensus among the observers in their assessment of the behavioural expression of the sheep. Observers described the Habituated and Lame sheep as significantly more focused/collected/assured compared to the Control sheep (P < 0.05). There was no difference in observer scores between the Inappetent sheep compared to the Controls. A number of associations were found between the QBA scores and the quantitative behavioural measures recorded. Sheep that baulked more frequently at the entrance to the WoW system (R-s = -0.70; P < 0.001) or had a greater number of circling events (R-s = -0.68; P < 0.001) were described as more reluctant/tense/wary, while those that recorded faster walking speeds (R-s = -0.65; P < 0.001) or spent less time standing stationary (R-s = -0.48; P < 0.01) were described as more focused/collected/assured. We conclude that qualitative and quantitative behavioural measures can be used to identify differences in animal behaviour, presumably related to their welfare state, when applied to short video clips automatically collected in a WoW setup. These findings suggest that behavioural measures could be collected, practically, within automated biometric data capture systems to provide additional information to aid in the assessment of sheep welfare in extensive systems.
