and tear over time. In reality, the problem
is even worse because, for feedstuffs, it is
difficult to maintain a set of samples over
long periods of time without changes
occurring in the samples.
The area of chemometrics dealing with
this subject has come to be known as
‘Calibration transfer’, or how to transfer (or
use) a calibration developed on one instru-
ment to another, or to the same instrument
after changes have occurred. Many proce-
dures have been tried and considerable
research is still being carried out to solve
this problem (Wang and Kolwalski, 1992).
The two central approaches used are: (i) to
make all the instruments as identical as
possible; and (ii) to alter the spectra so they
look as though they came off the same
instrument.
Approach number one is being used
on many Fourier transform NIR spectro-
meters (FTNIR). One of the big problems
with transferring calibrations is differences
in wavelength accuracy. For example, a
scanning monochromator may be designed
to collect data starting at wavelength
1100 nm. However, due to instrumental
variations, the actual wavelengths being
collected are off a little, so data point one
may be 1100.5 or 1099.5 nm. Only by
scanning a standard and comparing the
results with known values can the actual
wavelengths scanned be known. The
problem is that the wavelengths can be off
by varying degrees from wavelength to
wavelength. Thus, one might be able to
make 1100 exactly 1100, but then 1800
might be 1799.5 nm and 2498, 2498.5 nm,
etc. Instruments based on filters can have
similar problems due to the difficulty of
mass-producing identical filters. Fourier
transform spectrometers (Griffiths and de
Haseth, 1986) have the ability to lock in all
wavelengths accurately using a laser as the
standard. Since wavelength shift is one of
the biggest problems, some feel that
FTNIRs have an advantage with respect to
calibration transfer. At present, however,
their use has been virtually limited to
discriminate analysis of raw materials and
not for quantitative determinations of feed-
stuffs.
The general approach to making
spectra look alike has been to use regression
analysis to determine how the spectra vary
and to correct the variation. One weakness
with this technique is that a set of samples
similar in nature to those being tested is
needed for matching the spectra between
instruments. Even when using sealed
sample cups, it is difficult to produce
samples which do not change over time.
However, this appears to be one of the better
approaches at the present time. The proce-
dure is to run the set of standards periodic-
ally and then the software regresses the
spectra against a master set of spectra, finds
where differences exist and designs a regres-
sion equation to convert the new spectra to
the masters on a point by point basis. Using
this method, not only peak position, but
also changes in band shape and photometric
response (absorbances), etc. can be handled.
Once the equation is determined, the
spectra of all newly scanned samples can be
converted automatically. This procedure
can be used on one instrument, so that
spectra taken next year should look like
spectra taken today. It can also make several
instruments produce the same spectra for a
given set of samples. It is the latter which
has been used in networks of instruments.
The procedure has even been used to
convert spectra across vastly different types
of instruments. Recently, efforts have
concentrated on using a single sample to
perform the same transfer.
The biggest weakness with the proce-
dure mentioned above is its dependence on
a set of standard samples or sample. The
standards need to be similar to the types of
samples being tested since they correct for
band shape, etc. Efforts have been made to
seal forages in plastics, but these efforts
have not been successful. While it is a very
important subject, and research is continu-
ing, it is difficult to determine how much
success has been achieved to date.Instrumentation
Instrumentation for NIR comes in a wide
variety of sizes, shapes and forms. There198 J.B. Reeves III