Feed intake and efficiency are extremely important traits in the dairy industry, due to their impact on sustainability and profitability. Unfortunately, measuring these traits in commercial settings is difficult. Identification of indicator traits of feed intake or efficiency would be of great value. As sensor use on dairy farms has become more common, these technologies could be applied as proxies for feed intake, if first evaluated on research farms that possess feed intake measuring equipment. Thus, the objectives of this study were to assess the usability of automated sensor technologies as indicator traits for dairy cattle. Dry matter intake was adjusted for the energy sinks of milk and component yields and body weight, as well as contemporary group effects. Measures collected via sensor technologies were impacted by ambient temperature and health events (P < 0.10). The effect of THI and health varies by sensor type. Different health events impact the associations of sensor measures and feed intake for different durations of time. Fitting health at the time of event versus from the onset of illness until the end of the trial indicated that lameness has a longer lasting impact on feed intake and sensor measures (P < 0.05), compared to mastitis. Lame animals consumed more feed than healthy cohorts (0.58 – 1.66 kg/day; P < 0.10), whereas animals with mastitis consumed less (0.44 – 1.49 kg/day; P < 0.05). Absolute values of estimated effect sizes of sensor measures on ADMI ranged from 0.00003 to 4.90 kg/day (P < 0.05). The smallest estimate was for activity via ear tag one, without accounting for other factors. However, this measure was the most variable, making it potentially useful as a feed intake proxy. The largest estimated effect was for rumen temperature, when accounting for THI. This effect size is quite large; however, rumen temperature exhibits little variation and therefore has a similar impact compared to those with more moderate estimated effects. Significant interactions were observed between sensor measures and THI, as well as sensor measures and health events (P < 0.05). These interactions impact the interpretation of sensor associations with feed intake and are important to consider in implementing sensors as proxies of feed intake.

Two prediction methods were examined to determine if sensor measures could predict feed intake. The first utilized a single daily average sensor measure as a predictor of feed intake, whereas the second utilized multiple days at a time. Results did not indicate any predictive ability of these modeling types.

In conclusion, automated technologies may be useful for detecting illness and differences in feed intake in commercial farms. However, consideration needs to be given to THI and health status of the animal.