Windowing technique for determining the composition of organic samples by near-infrared reflectance

Abstract

A method of windowing regions of interest in near-infrared reflectance spectra is described. Windows are bounded by wavelengths at which spectra from all samples are re-scaled to zero. These bounds are determined numerically to maximize the correlation between the scaled spectral values within a window and chemical composition, as determined by wet-chemical methods, of a calibration set of samples. It is shown that the method yields predictions of chemical composition of forages that are, on average, more accurate than comparable methods based on principal components regression.