Sensor-based activity patterns of healthy calves housed in large groups

Summary

Young calves are susceptible to disease. Studies indicate that calf activity often changes prior to a clinical diagnosis. Accelerometers can monitor activity continuously, offering an opportunity for early detection of disease in individual reared calves, based on deviation from their ‘normal’ activity patterns. This requires the prior understanding of these ‘normal’ activity patterns in healthy calves. This study aimed at describing the group activity patterns of healthy group-housed calves. Holstein and crossbred calves (n = 231; 17 ± 4 d of age at arrival) were housed in six large pens (N = 38 ± 2 calves per pen). Calves had milk replacer via automated milk feeders twice or three times daily and ad libitum access to solid feed. High frequency activity data were continuously recorded from 6 to 28 weeks of age using leg accelerometers. Clinical examination was performed twice per week between 8 and 25 weeks of age, whereby any symptom of disease was scored based on the severity, and summed to reach a total ‘health score’. Activity data from sick calf days between two healthy days were extrapolated to surrounding ‘none-health monitoring’ days between the corresponding two healthy days. Generalized additive models with a Gaussian response were used to estimate daily group patterns of ‘being active’ and ‘being inactive’ per week, corrected for trends over time/age. Four features were extracted from the data and used in the model based on weekly averages: number of peaks, time (of the day) at which peaks occurred, the height (i.e. absolute value) of each peak, and the proportion of the night activity. The results showed that normal activity patterns can be described using the above features. The number of peaks in activity in a day went from 4 to 3 over the fattening period, with most peaks corresponding to availability times of new milk replacer. A peak in activity was consistently observed prior to darkness. Night-time activity was consistently around 20 % between 8 and 20 weeks of age and gradually increased to 27 % between 21 and 25 weeks of age. A leave-one-out analysis showed a medium accuracy (0.646) of using the fitted model to predict activity patterns of individual calves, indicating that most calves deviate in some way from this average pattern. The next steps are to identify which factors (e.g. personality, weight) lead to individual differences between calves in terms of activity and to develop valid models to detect deviations indicative of health issues in calves.