Quantification of climate impact for breeding and decision support at dairy farms (KB-46-005-020)

High-frequency behavioural sensor data of fourteen farms from the Netherlands and Belgium were collected in 2022. For all farms, cow information and hourly climatic data are also available. Using the individual cow behavioural time series and the relevant covariates (e.g. parity), we first construct biologically meaningful features by combining domain knowledge with high-frequency behavioural sensor data. We use these features to develop machine learning models that predict the individual cows’ sensitivity to heat stress. In step 1 we apply a univariate approach (impact on production and health separately), which is followed by a multivariate approach in step 2 (production and health combined). With the integration of multiple data sources (production, health, and behavioural data from the farms, and climatic data from weather stations), we further improve the current use of these data streams to support the adaptation of dairy farms to the changing climate. Upon better understanding and prediction of the impact of heat stress on dairy cows in different farm environments, farmers and breeders can (1) better anticipate on the impact of the stressor for individual cows, e.g. by separating the most vulnerable from the more tolerant animals, reducing its impact on production and the short and long-term welfare and health consequences; and (2) select the best animal-in-environment herd by identifying which cows deal better with the specific imposed (heat stress) challenges.