670 M. Dubernet
samples possible, but he/she must apply the most appropriate chemometric methods
and if possible, carry out critical comparative analyses of the results obtained by
these methods. Such an approach should always be validated by a rigorous statisti-
cal comparison of the results obtained using the selected model with measurements
obtained by classical analytical techniques. Several thousand determinations may
be required for this last step. Calibration of an FTIR instrument is therefore an
onerous undertaking which demands significant time and investment. To understand
better the difficulty of the approach, it is appropriate to specify the following critical
points.
12.5.5.1 Analyte Behaviour
Not all the organic compounds of wine behave in the same manner analytically for
several reasons.
First, it is clear that compounds at elevated concentrations (above 1g/L, for
instance) are likely to be more easily measured by virtue of the fact that they will
exhibit strong absorbance due to their concentration.
Second, the absorbance capacity of a compound depends on its molecular struc-
ture. Certain compounds, such as carbon dioxide absorb strongly in the mid-IR,
whereas other compounds, even those of substantial molecular complexity, have
less intensive absorbance.
Third, the influence of the other wine compounds (matrix effects) is far from
negligible. It is obvious that if the spectral bands of a target compound overlap with
those of water or ethanol, present in high proportions, the measurement sensitivity
for the compound in question will be severely compromised. Similarly, interactions
between constituents may lead to displacement of absorption bands.
It is easily acknowledged that here, again, theoretical knowledge is not of great
use and only an experimental approach can establish whether a compound can read-
ily be determined by FTIR. For instance, the determination of sugars is less reliable
for values less than 1g/L; it is evident that the more inaccessible the analyte, the
more complex the calibration and the more it needs to be executed with attention
and rigour.
12.5.5.2 Extrapolation
This is an essential point. In executing the calibration, it is not possible to assume
that extrapolation of the system is possible. Therefore, if the initial sampling does
not cover the entire analytical domain equally, the results obtained are less rich in
information and will thus be qualitatively deficient. By the same token, if there is
a significant matrix effect which arises in cases of matrices rich in polyphenols, it
is essential that the instrument is calibrated for wines of all probable polyphenol
concentrations at each concentration point for each analyte. If this is not the case
that part of the curve which is not calibrated for all possible phenol concentrations
will be more sensitive to matrix effects resulting in less reliable measurements.