1. the multispectral behaviour of the surface to be analysed; 
2. the characteristics of the measuring system. 
The data acquisition system only very seldom is fit thus as to measure the 
spectral properties in an optimum way. This leads to factors which combine 
high loadings of different surface properties in one factor and therefore 
can only very difficultly be interpreted. The solutions, in general obtained, 
are not the optimum ones for the analysis of causal interrelationships, 
demonstrated by the behaviour in the spectral bands: More evident features 
can be extracted by a rotation of the coordinate system. General procedures 
and methods, such as orthogonal and obligue rotation, exist. In many cases, 
it is not obvious by which optimizing principle the rotation is guided. As 
it is also possible to rotate the coordinate system by a visual rotation 
method, a few guide lines are listed here for orientation: 
1. The rotation should be maintained thus, that at least one factor 
loading of absolute low value (or the value zero) exists in each factor. 
2. For pairs of factors the loadings of some variables should be high in 
one factor and low in the other factor. 
3. There should be only a small number of variables of non-zero factor 
loadings in a pair of factors. 
As already has been pointed out, the rotation of the coordinate system leads 
to more easily interpretable factor loadings. For some case studies such a 
rotation is not necessary, as the clustering is around the principal axis of 
the first two factors. Sometimes the result of a rotation does not lead to an 
obvious improvement of the interpretability, so that the general question 
arises, if rotation is really a method to obtain reproducible results. One 
explanation for such a behaviour could be the above mentioned fact, that the 
measuring system is not fit to the problem to be measured, and therefore only 
secondary indicators of a physical effect are obtained. More investigations 
have to be made under this aspect.