Prediction of indigestible cell wall fraction of grass silage by near infrared reflectance spectroscopy

Samples of grass silages harvested at different maturities of primary growth (n=27) and regrowth (n=25) of timothy-meadow fescue swards, and samples from commercial dairy farms (n=42) were used to explore the potential of near infrared reflectance spectroscopy (NIRS) in predicting indigestible neutral detergent fibre (INDF) and digestible neutral detergent fibre (DNDF) content. The silages were analysed for cell wall (neutral detergent fibre, NDF) composition and incubated in situ for 12 days in nylon bags to determine INDF and DNDF. Dried and milled silage samples were scanned with a spectrophotometer over the wavelength range from 400 to 2498 nm. After mathematical treatments (standard normal variate transformation, detrending (SNV-D) and first-order derivatization), the spectra were correlated to NDF, INDF and DNDF of silages, and used to develop NIRS prediction equations by modified partial least squares technique. Cross validation was employed to test the robustness of NIR calibrations. As a comparison, several regression equations using chemical components as predictors were developed to predict silage INDF. The INDF content of grass silages was dependent on harvest, being higher for the regrowth than for the primary growth silages (83 g/kg dry matter (DM) versus 118 g/kg DM). Lignin content correlated to INDF, but a linear or non-linear regression equation based on lignin predicted INDF with a low accuracy (R2<0.4). The best chemical regression equation based on NDF and pepsin–cellulase solubility predicted INDF more accurately (prediction error 15.6 g/kg DM, R2=0.830). Evident correlation (r>0.5 or r<−0.5) peaks were observed between SNV-D spectra and INDF and DNDF close to wavelengths of 1260, 1450–1590, 1750, 2140 and 2420 nm, the regions at 1680, 1930, 2220 and 2280 nm being in common with lignin. Cross-validation statistics for INDF (prediction error 10.0 g/kg DM, R2=0.910) and DNDF (prediction error 19.1 g/kg DM, R2=0.823) showed that NIRS can potentially be used to predict digestion parameters with a precision that evidently is better than obtained with a regression technique based on any chemical parameters. Furthermore, since DNDF was assessed more accurately as a difference between NIRS estimates for NDF and INDF than with the straight NIRS calibration, future calibration development should be directed to improve INDF equations.