Prediction of organic matter digestibility in vivo of grass silage by near infrared reflectance spectroscopy: effect of calibration method, residual moisture and particle size

Abstract

The results of a study into the effects of calibration method, residual moisture content and sample particle size on the prediction of the organic matter digestibility in vivo of grass silages by near infrared reflectance spectroscopy are reported. A total of 182 grass silages with organic matter digestibility values measured with sheep were divided into calibration (103) and validation sets (79) on the basis of their spectral characteristics. Calibrations were developed using modified stepwise regression, modified partial least squares and principal components regression techniques with and without a scatter correction technique to reduce particle size effects. In addition, the use of a noise repeatability file during the calibration process was evaluated for its ability to reduce the effects of variation in sample residual moisture content. The highest predictive ability for organic matter digestibility was obtained using a second order derivative modified partial least squares regression with the use of both scatter correction and noise repeatability procedures. Validation statistics obtained were R 2 = 0.82, standard error of prediction 2.35 and slope 1.01. The best calibration was also shown to have acceptable analytical repeatability (2.44) and not to distinguish between silages made in clamps or as big bales wrapped in plastic.