Accuracy of in-line milk composition analysis with diffuse reflectance near-infrared spectroscopy

Knowledge of daily milk composition changes can assist in monitoring dairy cow health and can help to detect nutritional imbalances. An analytical tool offering the possibility of analyzing milk during the daily milking routine would provide such information. Near-infrared (NIR) spectroscopy can analyze multiple constituents in a given substrate at the same time. In this study, a special NIR in-line milk-analyzing device was designed, and its ability to predict the contents of fat, protein, lactose, and urea and the somatic cell count in milk during the milking process was evaluated. The NIR spectra were acquired with a diode array spectrometer in diffuse reflection in the wavelength range 851 to 1649 nm. The spectra originated from a total of 785 partial milkings out of 84 composite milkings. Corresponding subsamples of the composite milkings were used for reference analysis (n=785). Excellent validation results were obtained with regard to the coefficients of determination (R(2)=0.99, 0.98, and 0.92), and standard errors of prediction (0.09, 0.05, and 0.06) for fat (%), protein (%), and lactose (%), respectively. Satisfying results were achieved for urea content (mg/L) and logarithmically transformed SCC in milk, with R(2) of 0.82 and 0.85 and standard errors of prediction of 19.3 and 0.18, respectively. The accuracy of predicting protein, lactose, and urea contents was in accordance with international recommendations for reproducibility specified for in-line analytical devices.