Accuracy of milk composition analysis with near infrared spectroscopy in diffuse reflection mode

Knowledge of changes in daily milk composition can assist monitoring of dairy cow health and assist detection of nutritional imbalances. An analytical tool with the possibility to analyse milk during daily milking routine could provide that information. Near infrared spectroscopy (NIRS) can analyse multiple constituents in a given substrate at the same time. The accuracy of milk analysis with NIRS using diffuse reflectance for fat, protein, lactose and urea contents of unhomogenised milk was studied in this project. A total of 1610 of raw milk samples collected from 38 different cows were taken. Near infrared spectra in diffuse reflection in the wavelength range 851–1649 nm were taken. Calibration and validation for the different milk constituents were performed with chemometric tools. Satisfactory results were obtained with regard to coefficients of determination R2 = 0.998, 0.94, and 0.73 and standard error of prediction SEP = 0.03, 0.07, and 0.09 for fat (%), protein (%), and lactose (%) respectively. Poor results were achieved for predicting urea (mg l−1) with R2 = 0.31 and SEP = 40.23. The accuracy of predicted fat, protein, lactose and urea content in milk with NIRS was better or at least equal to similar studies in transmission or transflection mode. The NIRS analysis of raw milk constituents did not achieve laboratory accuracy but had the advantage that it can be done with higher frequency than the latter. Predictive accuracy was excellent with regard to fat and protein content in milk and adequate with regard to lactose content in milk when reproducibility recommendations for at-line devices were applied.