Farm Animal Metabolism and Nutrition

204 J.B. Reeves III

is known that the spectra found in the NIR
are due to overtone and combination bands
of absorptions found in the MIDIR (Murray
and Williams, 1987), the consequences of
the overlapping of the many bands present
in a multicomponent system are much less
understood. This overlapping may, for
example, account for the fact that NIRS
does not in general perform well for minor
constituents.

Minerals

One application area for which theoretical
concerns are known to place limits is
determination of minerals by NIRS. While
NIRS can be used to determine analytes
such as K, Na and P in forages, the bases
for these determinations are correlations
between the minerals and organic con-
stituents (Clark, 1989). This is because
these minerals do not have absorptions in
the NIR spectral region. Therefore, the
accuracy of mineral determination by NIRS
depends entirely on the degree of correla-
tion between organic constituents, which
do absorb in the NIR, and the mineral in
question. For example, it has been shown
that one can predict Ca, K, Mg and P from

the ADF and CP content of forages with an

accuracy equal to that achieved using NIRS

(Shenk et al., 1992).

High moisture samples

As shown earlier, NIRS does not perform as

well on high moisture samples as it does

for dried materials, even for the same assay

on the same samples. The question can

then be asked: are there fundamental

reasons why this is so and can they be

addressed? Research using simple systems

of water and single compounds, such as

sugars, amino acids, polysaccharides,

ketones, alcohols, amines, etc., has shown

(Reeves, 1993) that while as dry crystalline

solids, they give spectra which have sharp

peaks and which are unique to each sugar,

in solution, the spectra lack sharp peaks

and look much more alike, resembling

polysaccharides such as cellulose and

starch to a great degree. It was also found

that for ketones and alcohols in particular

that the addition of water caused peak

shifts, the degree of which was dependent

on the water concentration.

A second study (Reeves, 1994a) showed

that variations in pH, ionic strength and

Table 9.6.Accuracy of NIRS in analysing undried silagesa (Reeves et al., 1989).

Calibration set (n= 98) Validation set (n= 48)

Component R^2 SECb r^2 SEPc Biasd

Dry matter (DM) 0.97 2.07 0.95 3.15 0.05
Crude protein 0.97 1.02 0.96 1.25 0.11
Acid detergent fibre 0.91 2.59 0.90 2.78 0.70
Neutral detergent fibre 0.87 3.18 0.87 3.75 0.83
In vitrodigestible DM 0.87 2.39 0.73 3.78 0.28
Ammonia N 0.87 0.35 0.82 0.39 0.05
Hot water-insoluble N 0.95 0.68 0.82 1.28 0.42
ADF-insoluble N 0.71 0.65 0.59 1.40 0.31
pH 0.85 0.28 0.55 0.45 0.09
Acetic acid 0.74 0.56 0.57 0.79 0.05
Butyric acid 0.68 0.18 0.57 0.29 0.01
Lactic acid 0.71 1.41 0.74 1.30 0.01

aChemical composition expressed on a dry matter basis.
bStandard error of calibration.
cStandard error of performance.
dDeviation of NIRS mean from chemistry laboratory mean.