Quality evaluation of regional forage resources by means of near infrared reflectance spectroscopy

Abstract

Quality parameters of grassland and pasture samples collected during a three-year period at two environmentally and geographically different areas were analysed by Near Infrared Reflectance Spectroscopy (NIRS). Chemical analysis for crude protein (CP), crude fibre (CF), neutral detergent fibre (NDF), acid detergent fibre (ADF), acid detergent lignin (ADL) and crude ash (ASH) carried out on two-thirds of the samples were used in calibration processes. The remaining one-third of the data was used to validate the best calibrations obtained. Samples selection is discussed. Different math pre-treatments (derivative, gap, primary smoothing and secondary smoothing), light scattering correction methods and calibration algorithms were tested to achieve the better predictive performances. We obtained the best results using different regression algorithms to correlate spectral information to chemical data. For CP (R2 = 0.94, SEP=1.3), NDF (R2 = 0.95, SEP = 2.14) and ADF (R2 = 0.92, SEP=2.06) Multiple Linear Regression (MLR) models fit chemical data better than Mean Partial Least Square (MPLS) regression. A molecular basis explanation of wavelengths selected was carried out. MPLS models worked well for CF (R2 = 0.93, SEP=1.57), and ASH (R2 = 0.95, SEP=1.17) while poor calibrations were obtained for ADL using both algorithms. To confirm the reliability of the models developed, uncertainties of predictions were compared with findings on nutritional variations and animal performances.