of a spectral band. The combination of several high factor loadings in one 
factor indicate interrelations of spectral behaviour of these spectral bands 
caused by a physical effect which originated this spectral behaviour. The 
nature of the effect and its interrelationship to the spectral behaviour can 
very often not be very easily determined; it requires background knowledge 
for the explanation of the dependence of the physical or chemical effect and 
the spectral behaviour. This is the point where data analysis and user oriented 
experts have to cooperate. Factor analysis provides a quantitative measure as 
basis for the discovery, explanation and interpretation of interrelations which 
otherwise only qualitatively could be estimated. The information contributed 
by one factor to the overall information is identical to its variance contri- 
bution to the overall variance of the whole data set. 
As for certain case studies and for some application not all factors turn out 
to be necessary and useful, it is thus possible to determine the fraction of 
information extracted from the spectral data for a certain task. In general 
75 - 85 percent are sufficient and useful for evaluation purposes. The re- 
maining 25 - 15 percent in many cases can neither be explained nor do they show 
any indication to a physical phenomena. These factors mainly represent pheno- 
mena caused by sensor noise and measurement errors. 
For the extraction of the factors a variety of procedures have been developed 
[2] [3]. The easiest and best understood procedure is the one which bases on 
the principal component analysis. The main characteristics of the method is 
the fact, that each factor tends to maximize the amount of variance of the 
variables to be factored. Thus, the first factor represents the one with the 
highest variance. For the extraction of every following factor, the variance 
is intended to be maximized from the remaining amount of information. Thus, 
the information presented by every factor decreases with its sequence number. 
At the maximum, there can as many factors be extracted as there are variables. 
For the principal component analysis, the individual factors are uncorrelated, 
and a maximum of variance can be accounted with a minimum of factors. To make 
the method fully sufficient, it is necessary to use all available variables. 
The more of computer time has the advantage, that reliable new variables can 
be constructed, which allow from their position in the multidimensional factor