Impact of microclimatic conditions on dairy cows' milk production in an Automatic Milking System: evaluating trends and predictive modeling

Heat stress deriving from climate change represents a serious challenge to the dairy industry. Recent advancements in Artificial Intelligence have introduced a variety of tools capable of forecasting future conditions by utilizing past information as inputs, potentially offering valuable resources to address heat stress. This study aims to evaluate and predict the impact of microclimatic conditions on milk production within an Automatic Milking System (AMS). From 2016 to 2023 data on milk production and quality were collected from AMSs on a commercial dairy farm, while meteorological information was obtained from the closest ARPA (Agenzia Regionale per la Protezione Ambientale) station. Temperature and relative humidity data were combined to derive the Temperature Humidity Index (THI). Our analysis revealed a significant negative correlation between the protein content and THI over time, reaching a peak correlation of -0.66 with a lag of five days. This confirms a strong connection between milk quality and the surrounding temperature and humidity conditions. Consequently, understanding both current and historical microclimate conditions could facilitate predictions of future trends in milk-related data. Therefore, we employed a recent Neural Network (NN) model named TSMixer to forecast the protein, fat, milk production and milk temperature for the upcoming two months. This model utilizes input data from the preceding three months. TSMixer was trained and validated using data from 2016 to 2021, and subsequently tested with data from 2022 and 2023. The test outcomes demonstrate the NN's ability to accurately predict linear trends for the upcoming two months across protein, fat and milk production with R2 values of 0.83, 0.81, 0.80.