Comparison of the mid-infrared spectra and prediction equations developed from morning and evening milk samples from twice-a-day milked dairy cows

Abstract

Mid-infrared spectroscopy is a technology used globally for quantifying the concentration of fat, protein, lactose, and other constituents in the milk samples of both individual animals and bulk tank milk. Differences in the milk components and yield of cows are known to exist between morning and evening milk; nonetheless, differences in the spectra originating from the same cow from morning and evening milkings have never been investigated. Data were obtained from 2,602 dairy cows from 7 research farms in Ireland. A total of 199,288 morning milk spectra with associated evening milk spectra produced by the same cow within 24 h were available. Postediting, spectral data were available on the same 502 wavelengths in the mid-infrared region of the electromagnetic spectrum for all milk samples. Differences between morning and evening milk spectra produced by the same cow in a 24-h period were investigated using (1) the mean and SD of the difference between morning and evening spectra absorbance values, (2) the correlation between the morning and the respective evening wavelength absorbance values, and (3) the L₂ distance, all of which were quantified across stages of lactation, years, and farms. The average (SD) difference between the morning and the evening spectrum absorbance values produced by the same cow within 24 h was 0.00097 (0.008), and it was always larger than 0.055 for the wavelengths between 2,920 cm⁻¹ and 2,947 cm⁻¹. The correlation between morning wavelength absorbance values and the respective evening wavelength absorbance values were all strong (i.e., >0.80) in the spectral region of 1,469 cm⁻¹ to 1,473 cm⁻¹; weak correlations of <0.26 existed between morning and evening spectra wavelengths in the region of 1,593 cm⁻¹ to 1,597 cm⁻¹. These trends in correlations generally persisted within different stages of lactation, years, and farms. Results from the L₂ distance indicated that early lactation morning and evening spectra were more different from each other than when compared in late lactation; no large differences in the L₂ distance across different farms and years were evident. The impact of a prediction equation developed from morning spectral data but applied to evening milk spectral data, and vice versa, was investigated. Nitrogen use efficiency (NUE) was the animal trait explored; the mean NUE (SD) in the validation dataset was 22.16 (4.86). The root mean square error for predictions developed and validated on morning spectra samples was 3.49; this increased to 3.85 for the same validation in the morning spectra when the prediction equation was developed using only evening spectra. Similarly, the root mean square error from predictions developed and validated on evening spectra samples was 3.46, which increased to 3.85 when the prediction equation was developed using only morning spectra. In conclusion, morning and evening milk spectra produced by the same cow within 24 h differ, particularly in some spectral regions; these differences affect the prediction performance of applied prediction equations.