Prediction of the in vitro gas production dynamics of kikuyu grass by near-infrared reflectance spectroscopy using spectrally-structured sample populations

A study was carried out to test: (1) if the prediction of in vitro gas production of kikuyu grass (Pennisetum clandestinum) samples by near-infrared reflectance spectroscopy (NIRS) could be improved by the use of a spectrally-structured sample population; and (2) if the parameters from exponential models used to describe the kinetics of gas production could be calibrated by NIRS. Forty-one kikuyu grass samples (calibration set) were chosen out of a total of 288 on the basis of their spectral characteristics as representative samples of the whole sample population. Measurements of cumulative in vitro gas production were recorded at 3, 6, 12, 24, 48, 72 and 96 h. Spectra were transformed with 1st, 2nd or 3rd derivative mathematical treatments. NIRS calibration equations were derived for in vitro gas production, with or without scatter correction for particle size using modified partial least squares. The equations were validated using a set of 48 samples previously chosen at random from the total sample population (validation set). Satisfactory calibrations and cross-validations were obtained for all the static measurements of gas production (R2 = 0.77−0.86 (S.E. 0.48−2.06) and 0.74−0.82 (S.E. 0.50−2.18), for the gas volumes from 3 to 96 h, respectively), and the use of a spectrally-structured population improved the calibration and cross-validation statistics of the NIRS equations. However, when three exponential models were fitted to the gas production data, only the asymptote values could be satisfactorily calibrated or cross-validated. When the NIRS equations were used in the validation set, the static gas volumes were predicted with R2 values between 0.60 and 0.71 (S.E. 0.65 to 3.83, for the gas volumes from 3 to 96 h). These results were less accurate than within the calibration set, but they were still better than when the parameters from the exponential models were fitted. The results suggested that the calibration of static gas volumes is a more promising alternative than to fit specific parameters of kinetic fermentation models by NIRS