Modeling shade use of grazing dairy cows using sensor-derived data and machine learning

The health, welfare, and production impacts of heat stress are of increasing concern to dairy farmers. For cows grazing pasture, solar radiation is a key driver of heat stress load, and provision of shade (e.g., trees) is a primary mitigation option. To date, research on shade use choices by dairy cows has been limited to small-scale observation-based studies, and analysis at a commercial scale using sensor-derived data using machine learning has not been attempted. This study quantified the availability and use of shade at a herd level using cow global positioning system location data from herds of lactating dairy cows grazing pasture for 1 month in summer on 4 farms in the North Island of New Zealand, along with 15-min weather data and software that predicts potential shaded areas from aerial light detection and ranging imagery. Farms ranged from 60 to 400 ha of flat to rolling terrain. Cow shade use was quantified as the percentage of cows in shade above what would be expected from independent random walking and was modeled using 9 contrasting machine learning approaches. Shade use was found to be predictable and dependent on several interacting variables. Shade use increased when solar radiation exceeded 2 MJ/m² per hour, provided at least 1.3 m² of shade was available per cow, especially when air temperature exceeded 24°C and wind speed was less than 4.2 m/s. These results can be applied to historic or forecast weather data to help farmers make informed decisions about investment in shade resources or to maximize the use of existing resources.