Interferences

Interfering effects must be minimized. This is achieved by mathematical transformations, and two such accepted procedures are, first, the SNV (standard normal variate) transformation which standardizes the variance of the spectrum to unity with a mean of zero. This minimizes particle size effects and baseline drift. Second, the de-trending (D) transformation, which removes curvilinearity of the spectrum by use of a second-order polynomial correction (Barnes et al., 1989). Another procedure uses a repeatability file to remove interference from moisture on spectra from similar samples analysed over a long period (Deaville and Baker, 1993). The main spectral interference in the NIR of agricultural materials arises from the presence of water, which possesses a strong absorption at 1450 and 1930 nm. It is present even in materials dried at 100°C. Water forms strong hydrogen bonds with cellulose and other materials containing OH or NH groups. This makes the water difficult to remove and affects the absorption, causing a shift of up to 50 nm to longer wavelengths and band broadening (Shenk et al., 1992). The presence of moisture has two other quantitative effects: it will affect the height of the peaks and hence the estimated concentration of components; also, if the concentration is reported as percentage in DM, it will cause an underestimate of the true concentration.

To test whether it was practicable to relate any areas of the observed NIR

spectrum to chemical components, various species of grass and weeds were fed to sheep, and the strained rumen liquor and its protozoal and bacterial fractions were freeze-dried and the NIR spectra obtained. A typical spectrum for the ryegrass diet is shown in Fig. 9.1, the data being subjected to a standard normal variate and detrend procedure (SNV-D) before plotting against wavelength.

It proved impossible to relate the peak shapes and heights to a particular component and its concentration. The variation in peak shapes and heights are also extremely small, and visually difficult to discern. There are parts of the spectrum that are claimed to be areas of special biological significance. For the spectra to be meaningful, however, they must have been derived from a statistically significant number of similar samples to permit the use of correlation transform techniques to identify the analytically useful wavelength regions. These are not always available in investigative work. The margin between a valid statistical interpretation and doubtful conjecture is therefore very small in the case of the average analytical laboratory with only a limited number of samples with which to set up the algorithms. Only with organizations operating on a large scale does NIR gain in credibility. The interpretation of large numbers of NIR spectral data obtained from ADAS in relation to predicted and actual feeding value was the subject of a PhD thesis by Field (1995).

The capabilities of NIR to provide rapid analytical results that are acceptably accurate and precise, providing samples are similar to those used in the calibration procedure, are well established. It is still common practice, however, to analyse one sample in every 20 by wet chemical methods just to be sure that the calibration remains valid.

Fig. 9.1. The NIR spectra from strained and freeze-dried whole rumen liquor from sheep fed on ryegrass (solid line), and the bacterial (dotted line) and protozoal (dashed line) fractions. The standard normal variate and detrended data (SNV-DT) is plotted vs. wavelength. Log 1/R = - log (Reflectance) and is equivalent to absorbance.

Wavelength (nm)

Fig. 9.1. The NIR spectra from strained and freeze-dried whole rumen liquor from sheep fed on ryegrass (solid line), and the bacterial (dotted line) and protozoal (dashed line) fractions. The standard normal variate and detrended data (SNV-DT) is plotted vs. wavelength. Log 1/R = - log (Reflectance) and is equivalent to absorbance.

0 0

Post a comment