85 
7 
CULTURE / SAPLING I 
aspect of vegetation using the second scene of April. 
In the spring scenes the complete seperation of 
forest and non-forest was achieved by using the 
threshold method (definition of minima and maxima). 
The resulting forest mask was used as stratification 
for the subsequent classification of the TM-bands of 
the summer scene. Table 5 shows the selected minima 
and maxima in all the bands of the April scene. 
Using this threshold method it was also possible to 
seperate the signatures of the classes mixed deciduos 
forest and poplar from douglas firs, which overlap in 
the summer scene, since in the scene of April signi 
ficant difference could be recognized in the bands 
TM4 and TM5 for these classes. 
Using the digital elevation modell in the sub-montane 
landscape of the Kaiserstuhl the forests of the valley 
floor and that of the higher area were seperated by 
the definition of a height limit, which was set at 
200 metres: 
- Cn the one hand for forest areas with the same sig 
nature but needing a seperation for forestry purposes 
eg. mixed deciduos forest stands of the Rhine plain 
from mountainous mixed forest stands of the Kaiser 
stuhl could be seperated. 
- On the other hand the topographically related 
overlappings between signatures could be avoided eg. 
cultures and sapling stands of the valley from de 
ciduos forest with a southern exposition of the 
higher areas. 
All of the data sets were classified by the maximum 
likehood method. The wavelengths 1, 2 and 3 were in 
cluded for the SPOT-simulation, the wavelengths IMS 
2, 3, 4 and 5 for the TM-simulation. Similarly the 
Landsat 5/TM data set was classified with the bands 
TO 2, 3, 4 and 5 of the summer scene. The rejection 
threshold was given a value of 3 respectively 4 for 
the separating function, in order to reduce the num 
ber of unclassified pixels. 
The visuel comparison of the whole classification 
with the existing forestry maps showed a high degree 
of conformity. The quantitative verification is being 
undertaken at the moment. 
A digital forest map was used as overlay for the 
classification results of the SPOT- and TM-simulation 
data sets in order to estimate the average accuracy. 
The comparison showed that even areas of a size less 
than two hectares can be classified if signature 
differences are significant. 
analyst will rise considerably, partly because of the 
higher spatial resolution, but also owing to the new 
middle infrared wavelengths which is also planned 
for the coming SPOT-generations. 
5 CONCLUSIONS 
In an examination of the results it is important to 
consider the fact that the interpretations took place 
for limited test areas. The possibilities for use in 
larger forested areas are investigated at the moment. 
The high level of differentiation in the various 
bands, aimed for here, can cause problems if one con 
siders the high sensivity of the sensors to localised 
differences in site conditions. 
The choice of small training areas, caused partially 
by the size of the test areas in the research, would 
have to be extended by further training areas if a 
larger test area was used; it would also be impor 
tant to consider other factors,such as atmospherical 
influences. In hilly areas the effects of exposition 
can only be allowed for by using terrain models. 
The masking of deciduos and coniferous forest in 
multitemporal scenes is problematic if due to the 
geometrical rectification of different scenes the 
corresponding pixels don't match exactly. 
The evaluation of the SPOT-simulation for the 
forest areas in May gave just as good a classifica 
tion result as the TM-simulation in July, however by 
taking into account the middle infrared wavelengths, 
a more precise classification and more possibilities 
of differentiation was offered by the summer scenes. 
The thematic information for forestry areas will 
hardly be improved by the 20 metres resolution. 
The demands on the ground truth knowledge of the