A novel approach to estimate intake of lactating dairy cows through multiple on-cow accelerometer sensors.

Accurate prediction of intake of dairy cows will provide substantial improvements in herd health and management. Therefore, the objective was to evaluate the feasibility of using 3-dimensional accelerometer sensors for the estimation of individual intakes of lactating dairy cows. Twenty-four late-lactation Holstein dairy cows housed in a freestall barn were fitted with 3 sensors that record acceleration in the 3 axes (i.e., x, y, and z), one on the lateral side of the left hind leg and 2 attached to a halter directly superpose over the jaw and nose. Cows were assigned 2 groups, a data collection group (A; n = 12) and a validation group (B; n = 12). Cows were trained to use Calan gates during an adaptation period (7 d), and followed by 10 consecutive days of data collection of acceleration and individual intakes for both groups. Four cameras were used to continuously video record all cows, and eating times for each cow were generated. Sensors were set to record the 3D accelerations at 10-s intervals. Eating times and accelerometer data from group A was cross-reference based on date and time. Acceleration data corresponding to eating times and falling between the 1st and 3rd quartile was selected. The REG procedure of SAS 9.4 was used to regress the pre-selected acceleration data against the daily intake data for each cow to evaluate which acceleration combination could account for most of the varia J. Dairy Sci. Vol. 101, Suppl. 2 173 tion. The combination Y-nose + Y-jaw + Y-leg + X-leg had the highest R2 of 40.3%. The model obtained from this accelerometer combination in group A was tested in group B, and the overall estimated DMI was 23.2 ± 1.65 kg/d, while the actual DMI of group B was 23.7 ± 4.7 kg/d. Although the overall DMI estimation is accurate, a significant amount of cow variation is not accounted for in this model, and this is reflected in the greater standard deviation in the actual intakes in group B. These results suggest that multiple accelerometer data can accurately predict DMI; however, a more robust model will be obtained using a greater sample size