Prediction of the in vitro gas production and chemical composition of kikuyu grass by near-infrared reflectance spectroscopy

Abstract

The objective of this study was to predict the in vitro gas production and the estimated metabolisable energy (ME), crude protein (CP) and neutral detergent fibre (NDF) concentrations of kikuyu grass ( Pennisetum clandestinum) by near infrared reflectance spectroscopy (NIRS). A total of 288 samples collected in the Poas Region, Costa Rica were scanned (Population 1). The in vitro gas production and ME calibrations were done on a subset of samples in which gas production measurements (3, 6, 12, 24, 36, 48, 72 and 96 h incubations) had been previously carried out (Population 2) while 41 samples for the CP and NDF calibrations (Population 3) were selected on the basis of their H distances from Population 1. The parameters a, b, c and lag for the exponential equation p = a + b (1 − e − c( t − lag) ) (McDonald, 1981), where p is the volume of gas produced at time t, were fitted to the gas production data and an attempt was also made to predict them. The volumes of gas produced between 6 and 48 h were successfully calibrated and cross-validated. Coefficients of determination for the cross-validation (1 − RV) were 0.65, 0.74, 0.78, 0.70 and 0.60 for the volumes of gas produced at 6, 12, 24, 36 and 48 h respectively. The volumes of gas produced at 72 h could only be calibrated ( R 2 = 0.71) but not cross-validated, while the calibration results for the gas production at 3 and 96 h and the parameters for the exponential equation were poor. An analysis of the wavelength segments associated with the in vitro gas production indicated that the primary wavelength was always located between the 1664 and the 1696 nm spectral region regardless of incubation time. The estimated ME, CP and NDF concentrations were accurately calibrated and cross-validated. Standard errors of cross-validation of 0.23 MJ kg −1 DM, 11.4 g kg −1 DM and 15.9 g kg −1 DM were obtained for the ME, CP and NDF concentrations respectively. Scatter correction for particle size improved the performance of most of the equations across all constituents. The effects of different calibration methods, maths treatments and the factors affecting the results are discussed.