Evaluation of the IceTag leg sensor and its derivative models to predict behaviour, using beef cattle on rangeland

Background: There is interest in using animal-mounted sensors to provide the detailed timeline of domesticated ruminant behaviour on rangelands.

New method: Working with beef cattle, we evaluated the pedometer-like IceTag device (IceRobotics, Edinburgh, Scotland) that records step events, leg movement and body position (upright versus lying). We used partition analysis to compare behaviour as inferred from the device data with true behaviour as coded at high resolution from carefully synchronized video observations of 5-min duration.

Results: Malfunctions reduced the target dataset by 7%. The correspondence between IceTag and video-coded step counts was excellent (r²=0.97), and the device's indications of upright or lying corresponded well (error rate=1.4%) to the video-coded values. However, the proportion of steps that could be matched individually was relatively low (65% at a tolerance of 0.5s), and the indicated start of a lying bout was often triggered by leg movements of an upright animal. Partition analysis of Grazing versus Not-Grazing yielded an overall error rate of 22%. In both three- and four-way classifications of behaviour (Graze, Rest, Travel; Graze, Stand, Lie, Travel) error rates were low for non-graze behaviours, but only 25% of Graze observations were correctly classified; the overall error rate was 22%.

Comparison with existing method(s): The IceTag device performed well in mapping the diurnal patterns of animal position and step rate, but less well in separating grazing from upright resting.

Conclusions: Our results suggest that pedometry is not the ideal method for classifying behaviour when grazing is of paramount interest.