Farm Animal Metabolism and Nutrition

202 J.B. Reeves III

Chemical dependence

At present, NIRS is inherently limited by its
dependence on the calibration process, and
thus the chemically or physically deter-
mined calibration values. This leads to two
basic limits to the use of NIRS. First and
foremost, NIRS-determined values are on
average no more accurate than the chemic-
ally determined values upon which they are
based. The second and more subtle limit
relates to the basis of the chemical values.
Most of the chemistries used for animal
feeds are largely, if not completely, empirical
in nature. Therefore, measures such as NDF
and ADF determine whatever is insoluble
in that particular solution. This leads to the
possibility that only partial extraction of
components occurs. The question of how
this fractional extraction varies among
samples, and whether the fraction extracted
is spectrally the same or different from the
fraction not extracted, is one which has yet
to be answered. At present, the effect of the
empirical nature of many of the presently

used measures of composition and quality

on the accuracy and long-term stability of

calibrations is largely unknown. However, it

is generally felt that better chemical

analyses (more accurate and precise and

less empirical in nature) will lead to better

NIRS calibrations. Support for this exists in

the fact that NIRS calibrations are generally

best for dry matter and protein and decrease

in quality, in order, for measures of fibre,

digestibility and lignin. While lignin is in

fact a definable material composed of poly-

merized propenyl-benzene units, lignin as

measured can be a mixture of true lignin,

protein, waxes and carbohydrates, the exact

combination of which depends on the

sample in question and the procedure being

used (Reeves, 1993).

High moisture samples

While the previously noted effects also

apply to high moisture samples, these

materials have particular problems.

Table 9.4.Results of NIRS determination of study 2 in situsamples using calibration set from study 1a
(Reeves et al., 1991).

CSb VSc

Calibration Validation

Forage source source Assayd R^2 SECe R^2 SEPf Bias

Lucerne Second None DIG 0.99 1.77
(n= 64) CP 0.99 0.42

First Second DIG 0.69 8.95 28.32

CP 0.97 1.07 13.74

First+ Second DIG 0.94 3.59 0.95 3.41 0.01

CP 0.99 0.61 0.98 0.60 0.12

Orchard- Second None DIG 0.99 1.69
grass CP 0.98 0.46
(n= 64) First Second DIG 0.94 4.52 72.35

CP 0.97 0.98 1.28

First+ Second DIG 0.98 2.45 0.98 2.91 0.70

CP 1.00 0.26 0.98 0.61 0.13

aStudy 2 samples digested and analysed at the University of Maryland, study 1 at Beltsville.
bCS = calibration set: second = all study 2 samples, first = all study 1 samples, First+ = all of the study 1

samples plus six of lucerne and/or orchardgrass samples from study 2 as appropriate.
cValidation set: second = all samples from study 2.
dDIG = percentage digested, CP = crude protein.
eSEC = standard error of calibration.
fSEP = standard error of performance.