Potential of Individual Cow Scatter Correction for an Improved Accuracy of near Infrared Milk Composition Analysis

Abstract

Near infrared (NIR) spectroscopy has the potential to quickly analyse raw milk composition on the farm. The accurate determination of individual cow's milk composition changes can indicate health imbalances and, hence, support herd management. The accuracy of NIR calibration models are highly influenced by individual cow scatter effects in the NIR spectra based, to a large extent, on individual scattering effects of milk fat globules (MFG). In this study the potential of an individual cow scatter correction was investigated. A total of 1151 milk samples were taken from 12 Holstein cows and used for spectra acquired in the 851–1649 nm wavelength range with a diode array spectrometer. Corresponding milk samples from the composite milkings were used for reference analysis. Individual cow scatter correction factors were developed from raw milk samples and homogenised milk samples. Calibration models based on the partial least squares (PLS)-regression method from corrected and uncorrected spectra were compared with regard to their prediction accuracy of fat (%), protein (%), lactose (%) and urea content (mgL−1) as well as logarithmised somatic cell count (log SCC) in milk. Prediction of fat content in raw milk was excellent both with and without scatter correction. Improved calibration results were obtained particularly for predicting the content of protein, lactose, urea and log SCC in raw milk when calibration models were based on the corrected spectra. Compared to prediction results of the uncorrected dataset, root mean square error was reduced by up to 25%.