An AI-based hybrid model for dairy cattle heat tolerance phenotype

Extreme weather events pose substantial risks to dairy cattle production and welfare, with heat stress being a major disruptor. Current approaches for heat tolerance phenotype selection in dairy cattle primarily utilise statistical models. We propose a novel hybrid AI-based method (HAIM) to enhance the selection of heat-tolerant cattle. It integrates AI with a statistical genetics model that links heat tolerance to decline in milk yield as the Temperature-Humidity Index (THI) rises above 60. This decline is modelled at both herd and individual levels, with the individual decline per unit increase in THI represented by the random slope and the intercept providing the baseline. For this modelling, 20 years of climate data were combined with New South Wales Holstein dairy cow production data, covering the same period and resulting in 3 million records. The HAIM model demonstrated a smaller standard deviation of residuals (3.25 L/cow/test day) compared to the statistical model (3.37 L/cow/test day). A high correlation (R² = 0.93) between the random slopes estimated by both models indicated similar heat tolerance estimates, with the AI-based model showing lower standard deviation of residuals. A negative association between estimated random intercepts and slopes highlighted that high-producing animals are more vulnerable to heat stress. These findings emphasise the importance of considering both traits (random slopes and intercepts) to improve heat tolerance selection without compromising overall production levels. The proposed model combines the strengths of statistical modelling with the capabilities of AI, resulting in a more accurate and efficient means of identifying heat-tolerant cattle.