detection and identification for either coniferous and decideous trees. In 
contrary to the example of the decideous trees no significant clustering 
occured for the day 160. For coniferous forest clustering only occured for 
the day 182 but not so significantly as it used to be for the decideous 
forest type. The infrared spectral behaviour only changed slightly in contrary 
to be behaviour of the decideous forest area. 
Example 6: Marsh Area (FMP) (Fig. 6) 
  
On day 160 vegetation in form of dry grass and dry small bushes covered the 
soil in the marsh area. The water contents of the soil surface was very low. 
After a rain period in June 1976 green vegetation dominated in the area. The 
change in the spectral interrelationship can be observed from the factor 
loadings for these two days. A comparison to the results of the decideous 
trees show, that clustering also occures for the channels 3, 4, 5 and 7 as 
well as for channel 8, 9 and 10. The location of the latter in factor pattern 
space is, however, different. Also of interest is the clustering of the 
channels 1 and 2. 
These examples should be considered as demonstration of the application of 
factor analysis for data evaluation in order to obtain indicators of inter- 
relational behaviour of different spectral channels. In order to fully make 
advantage of the information content which in this form is not hidden any 
more in a correlation or covariance matrix, the colaboration between data 
evaluator and user oriented specialist is mandatory. Otherwise, factor 
analysis could become another method where new numbers are generated from old 
ones without deriving new insight in the effects of physical phenomena to the 
spectral behaviour of natural and artificial surfaces. Only thus, by the 
explanation of the causal interrelationship between spectral behaviour and 
physical it is possible to reduce the volume of data necessary to be evalua- 
ted for earth scientific remote sensign applications. 
4. Significance Tests on the Number of Factors 
  
The result of a factor analysis is as good as the data base is. For this 
reason, a significance test on the correlation matrix on the principal compo- 
; ; 2 ; 
nents is mandatory. For either one of them, the x -test of Bartlett is