Estimating daily fat percentage from single samples in herds with automatic milking system using a regression model

Abstract

Constant access to the automatic milking system (AMS) leads to varying milking frequency of cows and subsequently varying milking interval lengths (MI) and milk yield (MY) of single milkings. This influences milk production and can result in variable milk composition in individual milkings during the day.

In this study, different effects of animal and milking frequency associated characteristics on the milk components fat and protein from herds milked with AMS were explored and analyzed with respect to their influence on daily content. From this, a new regression model was developed to enable the estimation of daily fat percentage from single sampled milkings on test day. Collectively, 909,922 test day records from 176,926 cows in Germany and Austria milked two to three times within 24 h with AMS were assembled. As the complex process of fat production in the udder is notably affected by MI and MY, common regression models address this issue by mainly relying on fat%, MY and MI of two preceding milkings. The new regression model increases the number of included preceding milkings up to four covering a timespan of >24 h. Moreover, the model is complemented by lactation stage, parity, and daytime of the sampled milking. Its performance was compared to those of two regression models from literature. With a root mean squared error of 0.275, a mean absolute error of 0.195 and R² = 0.816 the new model outperformed these reference models. For 62.6 % of test day records, the developed model estimated a more precise daily fat content than the measured fat content from a single milking.

Therefore, this model is an adequate solution for estimating the daily fat content for AMS herds, which are not able to take more than one sample on test day.