selection of the most important spectral channels for discrimination and 
  
    
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
     
explanation of physical effects is difficult to maintain. 
The purpose of this paper is to introduce a method which best selects the 
significant spectral channels of a surface for the quantitative description 
of the spectral behaviour and their changes in time and space, and to dervive 
from these data an indication of the physical phenomena which are responsible 
for this behaviour. The type of surface and object considered is assumed to be 
known already prior to the analysis. The channels to be selected for this pur- 
pose are not necessarily the same as those for the discrimination to other 
surfaces. An intensive literature study to collect some experience for spectral 
channel selection to derive the responsible physical phenomena for the spectral 
behaviour, did not result in reproducible indicators. The reasons are manifold: 
1. The relative radiation intensities of a surface in different spectral 
channels do not necessarily represent the real (absolute) radiation con- 
ditions, because the intensity calibration window for the individual 
spectral bands is shifted to obtain an extrem dynamic range for the image 
display and not for the sensor sensitivity. A selection of the most 
significant spectral channels thus in most cases is maintained in such a 
way as to best distinguish a surface from its surrounding by its image 
display contrast. This, however, does not necessarily result in the 
selection of the most important spectral channels for the derivation of 
physical phenomena. 
2. Different authors do not talk about identical surfaces and phenomena when 
they use the same terms. A typical example is here the feature "forest", 
which shows a great variety in the form of appearance. Thus experience 
gained for one surface cannot be generalized and shows therefore a low degree 
of reproduction. 
3. Spectral and spatial resolution are not necessarily compatible and thus not 
comparable for different surface types and data acquisition methods. Thus, 
the result of spectral channels selection depends particularly on the data 
acquisition system. The descrepancy can be caused here by different bandwidth 
and positioning of their central wave length of the individual spectral 
channels.