402 
3.2 Separation of Forest Areas 
Forest areas were separated from 
non-forest areas by applying a threshold 
method. A forest mask is produced on the 
basis of this method, using several TM 
bands of two seasons (Keil et al, 1988, 
1990). Similar methods were used by 
Schardt (1988). By including the second 
season the accuracy in the delimitation 
from agricultural areas is considerably 
increased. 
The method had to be extended in order 
to integrate young cultures and clea 
rings in the forest, which could not be 
separated from agriculture merely by 
spectral signature. Using raised 
thresholds resulted in additional areas 
with young cultures, clearings, as well 
as agricultural areas, however; these 
areas were compared with digitized 
forest layers of the related topographic 
maps 1:100 000 and corrected, using this 
additional information (for detailed 
description see Keil et al, 1990). 
3.3 Selection of Training Areas and 
Signature Analysis 
About 600 training areas were recorded 
for the three map sheets to serve as a 
basis for the supervised forest 
classification. The selection of trai 
ning areas was performed by ground truth 
assessments, in connection with 
available aerial photographs and forest 
management information. Above all the 
following stand parameters with their 
effects on spectral signatures were 
studied more closely: 
. Tree species and natural age classes 
. Types of mixtures and proportions of 
mixing 
. Canopy density, underbrush, and ground 
cover 
. Topographical features, such as expo 
sition and inclination 
. Any damage due to storm or snow-break 
The division of the study area into 
growth zones and growth districts gave 
valuable assistance in distributing the 
training areas and preparing the 
classification. For each growth 
zone/growth district, stands which were 
as representative as possible were 
assessed in order to be able to compare 
signatures within the growth districts 
as well as between the various growth 
districts. 
The comparison was based on the analysis 
of spectral signatures of the training 
areas. Examples of signature plots of 
different stand types are shown in Fig. 
3 and 4. A detailed discussion on the 
results of spectral signatures was 
performed by Keil et al., 1990. 
In particular in the context of a large 
area mapping, it is important to regi 
ster the regional differences in 
spectral signatures. These differences 
are partly very serious, above all for 
Fig. 3: Spectral signatures of spruce 
and pine stands 
Fig. 4: Spectral signatures of mixed 
spruce and beech stands of 
different mixture percentages