Modelling of heat stress in a robotic dairy farm. Part 2: Identifying the specific thresholds with production factors.

Thresholds of heat stress are identified by determining the values of thermal comfort indices with significant change of animal responses. However, published thresholds may lead to inaccuracy when dealing with specific climate conditions, animal breeds and production factors. Thus, determining dynamic thresholds might provide better assessment of heat stress, with self-calibration capabilities. In this study, a large dataset of individual age, body mass (BM), days in milk (DIM), daily milk yield (DMY) and milk temperature (MT) of 126 lactating Holstein cows was collected from a robotic dairy farm over five years. The ambient temperature data was collected from a local weather station and processed as daily minimum and mean temperature (T min and T mean). For the whole herd, a new series of heat stress thresholds with stages were defined as comfort stage, milk heat stress, effective heat stress and critical heat stress. The definition was based on the cow's responses in DMY and MT, which provides a potential approach to accurately alert for heat stress in robotic farming systems by using the existing data source. For the specific individuals, dynamic thresholds of heat stress were identified and categorised using the decision tree machine learning model. The categorisation achieved 79–94% overall accuracy, and demonstrated the importance of cooling cows during their early lactation period. • Use broken-line regression to assess cow's heat stress link to production factors. • Decision tree model to choose dynamic threshold of heat stress for individual cow. • Identify multi-thresholds indicating none, mild, effective and critical heat stress. • Demonstrate the importance of cooling cows during early lactation stage. [ABSTRACT FROM AUTHOR]