Abstract

The objectives of this study were to determine the advantages and disadvantages of using mid-infrared diffuse reflectance spectroscopy versus near infrared for the analysis of diversely-treated forages and by-products, and to compare the relative merits of calibration development using multi-linear regression versus partial least squares regression for such samples. Study samples (325) were randomly selected from 2808 possible sample–treatment combinations generated using 18 feedstuffs, 26 treatments (acids, bases, oxidising and reducing agents, heat etc.), and six levels of reagent for each treatment. Neutral detergent fibre and crude protein were determined on all samples and calculated on an organic dry matter basis. Neat samples were scanned (256 co-added scans) from 4000 to 400 cm⁻¹ at 4 and 16 cm⁻¹ resolutions on a Fourier transform infrared spectrometer equipped with a sample transport device, and in the near infrared (1100–2498 nm) using a scanning monochromator (64 co-added scans). Calibration results using 4 cm⁻¹ resolution mid-infrared spectra were equal to those obtained using near infrared spectra. While the results for both spectral regions using partial least squares regression were an improvement over those using multi-linear regression, neither region produced calibrations equal to those obtained for less diversely treated or non-treated samples. The results obtained also showed that: (1) visual determination of likely compositional or spectral outliers is not obviously feasible using either near or mid-infrared spectra; (2) changes in spectra due to treatment can be much more easily attributed to compositional components using mid- as opposed to near infrared spectra; (3) examination of either mid-infrared PLS or PCR factors can be useful in determining compositional changes caused by treatments, while near infrared PLS and PCR factors appear to be dominated by moisture differences.

© 1999 NIR Publications