Farm Animal Metabolism and Nutrition

is tilting from left to right, also resulting in
incorrect absorbance values, but, in this
case, the error varies in intensity from 0 on
the left to a maximum on the right side of
the peak. This type of error is called multi-
plicative scatter (Isaksson and Kowalski,
1993) and is also common among ground
feed samples due to particle size and
absorbance differences, but requires more
than a simple baseline shift for correction.
In reality, as opposed to this simplification,
multiplicative scatter is a function of the
absorbance values (i.e. the greater the
absorbance the larger the shift). Finally, in
Fig. 9.4E, we see the result when a second
component is present (F), and the resulting
spectrum E is a combination of the two
component spectra (A and F). In this case,
using the absorption at the peak also gives
incorrect results. This could occur if, for
example, a sample was tested using a cali-
bration built from samples which had no
component F. Since no F was ever present
in the calibration set, there would be no
compensation built into the calibration for
it. This, combined with the fact that virtu-

ally anything different about samples

(physical or compositional) or the condi-

tions under which spectra were obtained

(instrumental, environmental, etc.) can

alter NIR spectra and thereby influence the

accuracy of a calibration, probably explains

more about the nature and problems of NIR

calibration development, maintenance and

use than anything else. Since NIR calibra-

tions often use many wavelengths, one can

imagine the total effect of many such spec-

tral alterations on a calibration.

Although developing a calibration or

equation to determine the composition of

feed samples from their NIR spectra, one is

trying to find relationships between the

spectra obtained and the component of

interest (NDF, ADF, etc.) in order to deter-

mine the composition of new samples

using only their spectra. Therefore, any-

thing which is different between the

spectra used to develop the calibration

initially, and the spectra of future samples,

which is not due only to compositional

variations which have been included in the

calibration, can result in errors. A list of

Use of Near Infrared Reflectance Spectroscopy 189

Absorbance

1000 1200 1400 1600 1800 2000 2200 2400

Wavelength (nm)

Multipoint baseline corrected

Cellulose

Casein

Fig. 9.3.Baseline-corrected NIR spectra of cellulose and casein.