Discriminating spontaneous locomotor play of dairy calves using accelerometers

Play behavior is a promising welfare indicator in dairy calves because it decreases in negative situations such as pain or hunger and increases in positive contexts such as in appropriate social environments. Directly measuring play is time consuming because it is performed in irregular bouts and can be inconsistent over days. To facilitate automatic recording of play, previous studies fitted triaxial accelerometers to the hind legs of calves and measured the velocity of movements in large arenas; high correlations were reported between vertical axis peak duration and the duration of locomotor play. The current study aimed to validate accelerometers for recording spontaneous locomotor play in calves' home pens over longer periods. Data were collected from 48 Holstein Friesian calves, housed in groups of 3 in pens of 10 m(2), at either 4 or 8 wk of age. Acceleration at the vertical axis of the hind leg was recorded at a rate of 1 Hz. One active time period for each calf was randomly selected (mean duration +/- standard deviation = 34 +/- 9 min). From video of the corresponding time period, the frequency of locomotor play events, consisting of run, turn, and buck/buck-kick, was recorded using behavior sampling. Combined counts of play events were highly correlated (Pearson r = 0.91) with counts of acceleration peaks. However, for calves with higher levels of locomotor play, this method underestimated the extent of play. Alternatively, run, turn, and buck events obtained from video were transformed by creating intervals of 10 s and then classifying each 10-s interval as comprising events of play (play) or not comprising events of play (no play). The corresponding accelerometer data for all 10-s periods, equaling 10 consecutive readings each, were classified into play or no play by using quadratic discriminant analysis; 79% of periods with locomotor play were correctly classified. Counts of observed play intervals correlated with the counts of play periods from accelerometers (r = 0.87), but the discriminant analysis consistently overestimated play. In conclusion, accelerometer measurements at 1 Hz (in 1-s intervals) and at the vertical axis cannot be used alone to exactly quantify absolute levels of locomotor play in the home pen. However, counts of peak accelerations can provide a rough estimate of inter-individual differences in play events, and discriminant analysis can be used as a proxy for one-zero sampling of inter-individual differences in locomotor play.