2 
Lon 
rvive 
sible 
to be 
3 pur- 
ar 
jectral 
jectral 
ılfold: 
nage 
1 à 
ww 
of 
us not 
Thus, 
data 
andwidth 
al 
4. While describing the spectral characteristic of different surfaces the 
location and the seasonal variation very often is not taken into conside- 
ration and indicated. 
In order to really select the most important spectral channels which also by 
their interrelational behaviour significantly indicate physical or chemical 
phenomena, all spectral channels available have to be considered, even those 
which are highly correlated. High correlation sometimes presents hidden infor- 
mation which otherwise would be lost if either one of these the linearly highly 
depending spectral channels is ignored. The basis for the selection method 
therefore will be the correlation or covariance matrix instead of the individual 
intensity values and their variation. 
A selection of the most important spectral channels which best describe the 
physical reason of a phenomena is only possible to a certain extent with the 
data available because the bandwidth and the central frequency of sensitivity 
of every spectral channel is already fixed in discrete instead in a continuous 
form. This means that an optimum selection of spectral bands by varying centre 
position and bandwidth, as it would be possible for the continuous form, cannot 
be maintained and therefore, an interrelationship between physical effect and 
spectral phenomena cannot always be derived, because different physical pheno- 
mena appear as events in different bandwidth. Fixed positioning of spectral 
bands fades out slight spectral changes; the occurence of a physical phenomena 
is not measured. 
2. Principles of Factor Analysis 
  
Factor analysis is a generic term for a variety of procedures which were de- 
veloped for the purpose of analysing interrelations within a set of variables. 
For the reasons mentioned above the correlation or covariance matrix is used 
as data base. The intention of the factor analysis is to determine a set of 
vectors whose product again generates the original correlation or covariance 
matrix. The vectors thus obtained are called factors and the components of 
every factor are called factor loadings. The higher the absolute value of a 
factor loading within a factor is, the more important is the information 
provided by the variable concerned, in our case the information contribution