International Archives of Photogrammetry and Remote Sensing, Vol. 32, Part 7-4-3 W6, Valladolid, Spain, 3-4 June, 1999 
likelihood classifications with maximum likelihood tie 
resolution and three standard deviations were used. 
Training areas have been selected using as sole criterion the 
human capability to discriminate among different colours of 
each FCC. As an additional help to the classification process, 
signature separability reports and scatterplots were used. A 
great effort was made in order to create as many representative 
classes for each approach as possible, so that the majority of the 
pixels were classified as one of the available classes. It was not 
always possible to use the same training areas for the 
multispectral and the anisotropy approaches. The majority of 
the classes created were the same, but the total class number and 
the described objects varied a little. 
Classification accuracy was assessed by comparing about 100 
classified pixels with management maps of the Diimast 
experimental farm and forest management maps. No 
information was available for the privately owned farmland. 
4. RESULTS 
Using visual interpretation, only the complementarity 
capabilities can be estimated. For evaluating the synergy effects, 
a combined analysis of all mode D bands has to be performed. 
4.1. Visual interpretation 
The large landcover classes forest, agricultural areas, settlement 
areas, water bodies and infrastructure are detectable with similar 
accuracy in both FCCs. Nevertheless, due to the „structured“ 
impression of the „anisotropy“ FCC, the visual delineation of 
settlements seems to be easier. The further differentiation of 
these main landcover categories shows object-specific 
advantages for the one or the other approach. 
Fig. 9. Forest type discrimination: left „anisotropy“, right 
„multispectral“ approach. 
As far as the discrimination between the forest types is 
concerned, the anisotropy approach does not provide a clear 
visual discrimination. The multispectral approach provides 
better results in this respect (Figure 9). However, in the 
multispectral approach the same cultivation (wheat) appears 
with different optical characteristics within a distance of a few 
km. The anisotropy approach appears to have no problem 
depicting the same cultivation with similar optical 
characteristics in both areas. 
Comparing the multispectral and the anisotropy approaches, the 
latter one describes and discriminates more accurately the 
boundaries of agricultural landcover types (Figure 10). The 
capability of discrimination of crop types is high in both 
approaches (see 1 (rye), 2 (winter wheat), 3 (triticale) and 4 
(maize) in Figure 10). Different features in fields 5, 6 and 7 
demonstrate that each approach displays specific characteristics 
of the object of interest, which may be used for further 
discrimination. 
Fig. 10. Top: „Anisot“ FCC; bottom: „MS“ FCC. Separability 
of crops: rye (1), winter wheat (2), triticale (3), plot 
with maize seed (4). Plots 5, 6, 7 show different 
features in the two images. 
Meadows in the wet and marshy area of the Munich gravel 
plane at the border to the tertiary hills are partially mowed. In 
the MS FCC fresh mowed plots look like bare soil. In the anisot 
FCC, the same plots have the signature of sparse vegetation, 
describing the real situation much better. 
4.2. Classification results 
The anisotropy approach showed better results than MSS, as far 
as the assignment of the pixels to the available classes is 
concerned. The combined approach, however, had the best 
overall performance and accuracy. 
The anisotropy approach managed to distinguish wheat as one 
class in the whole scene with high accuracy results. The other 
two approaches failed and classified wheat in two different 
classes. Moreover, rye and triticale, a cross-breeding between 
wheat and rye, could be differentiated by the anisotropy 
approach, while in the two other approaches both rye and 
triticale were classified as one class. 
The class that represented urban/man-made constructions was 
more accurate in the multispectral approach, but often 
misinterpreted as coniferous forest in the combined approach. 
The latter phenomenon appeared as a buffer zone at the 
boundaries of this forest class with other ones. It also appeared 
to a high degree in the anisotropy approach, where it could be 
noticed in other areas as well. One possible interpretation is that 
this phenomenon has to do with the shadow effect of the