Grazing time of dairy cows recorded using two devices based on accelerometers

International audience

English. Investigating grazing behaviour of dairy cattle requires accurately recording their daily grazing time. Unfortunately, commercially available devices are often not validated for the degree of precision required in animal science at the cow x day level. In two studies, the ability of the Feed'Live (R) device (New Medria) to accurately estimate daily grazing time of dairy cows was investigated. It contains a triaxial accelerometer whose raw data, once processed by New Medria, provides the dominant activity for 5-min periods: intake, rumination, over- activity or other activity. Feed'Live data were compared to those from the Lifecorder Plus (Kenz) device, which contains a uniaxial accelerometer, used in the study as reference values for actual grazing time due to its known high accuracy thanks to a large external validation. Twenty four dairy cows were simultaneously equipped with both devices for several weeks in study 1 (spring 2021) and study 2 (spring 2022), for a total of 1224 cow x day datapoints. Daily access time to pasture was either 19 h/day (861 cow x day datapoints) or 7 h/day (363 cow x day datapoints) according to the experimental treatments investigated. On the whole database, compared to the Lifecorder Plus, Feed'Live underestimated daily grazing time by 15 % (65 min/day), with a lower underestimation for cows having 19 h/day of access to pasture (11 %, 48 min/day), and a greater underestimation for cows having only 7 h/day of access to pasture (31 %, 107 min/day). The bias in the estimated grazing time between the two devices was strongly and positively correlated with the time of other activity. This is because Feed'Live failed to identify grazing as the main activity during several grazing sequences and sometimes missed entire grazing sequences. Most of the grazing activity that the Feed'Live failed to detect was classified as other activity, with no misclassification as rumination. Correcting grazing time by adding other activity time to it provided a much more accurate estimate of actual grazing time, with no more bias and an average mean prediction error of only 9 % at the cow x day level vs. 23 % without correction on the entire database. For cows having access to pasture only during daytime, the mean prediction error was reduced from 40 % to 6 % thanks to this correction. For scientific purposes, when data are available, actual grazing time can thus be estimated accurately by the Feed'Live by summing the times of grazing and other activity.