Prediction of selected ewe's milk properties and differentiating between pasture and box feeding using visible and near infrared spectroscopy

Abstract

Rapid measurement of milk properties and discrimination of milk origin is essential for quality control of milk products. Visible (Vis) and near infrared (NIR) spectroscopy is a proven technology to provide intensive, cost effective and fast analysis with high accuracy for many materials. This study was undertaken to explore the potential of Vis–NIR spectroscopy to measure ash, density, fat, freezing point, lactose, non-fat in dry matter, pH and protein of 50 Tunisian ewe's milk samples from different feeding systems and genotypes. These samples were milked from Comisana genotype with pasture feeding (CO-P), Sicilo-Sarde genotype with pasture feeding (SS-P), Sicilo-Sarde genotype feeding on Scotch horse bean meal (SS-B) and Sicilo-Sarde genotype feeding on soybean meal (SS-S). Two spectrophotometers with different detectors and wavelength range were compared for accuracy of measurement. A mobile, fibre-type spectrophotometer (Zeiss Corona 45 visnir fibre, Germany) had a short Vis–NIR wavelength range of 300–1710 nm and consisted of two diode-array detectors, whereas a mobile spectrophotometer (LabSpec@Pro Near Infrared Analyser, Analytical Spectral Devices (ASD), Inc, USA) had a full wavelength range of 350–2500 nm and consisted of one diode-array and two monochromator detectors. Milk samples were divided into calibration (40 samples) and validation (10 samples) sets, and the partial least squares regression (PLSR) with leave-one-out cross-validation was adopted to develop calibration models of different milk properties. Results obtained with both spectrophotometers showed that PLSR differentiated between milk samples from different feeding systems (pasture or box feeding). Using the validation set, pH and fat were poorly predicted ( R 2 0.14–0.73 and ratio prediction deviation (RPD) between 1.11 and 1.64), whereas the remaining properties were predicted with good accuracy ( R 2 0.78–0.90; RPD 2.20–3.34). The good accuracy models developed with the ASD instrument outperformed ( R 2 0.84–0.90; RPD 2.34–3.34) the corresponding ones developed with the Zeiss instrument ( R 2 0.78–0.81; RPD 2.20–2.41). However, the Zeiss instrument provided better accuracy for the properties that were difficult to measure, e.g., pH and fat. Therefore, it is recommended to adopt a full wavelength range Vis–NIR instrument for good accuracy measurement of key properties in ewe's milk of different feeding systems and genotypes.