406 
Spruce 
Mixed 
Spruce/ 
Pine 
Pine 
Mixed 
Decid./ 
Conif. 
Deciduous 
Clearings/ 
Cultures 
Nonforest 
Sum of 
Classified 
Pixels 
Spruce 
74.8 
18.1 
3.8 
2.3 
0.0 
0.5 
0.3 
5018 
Mixed 
Spruce/ 
Pine 
21.1 
60.0 
18.1 
0.5 
0.0 
0.0 
0.0 
3238 
Pine 
8.0 
9.8 
80.0 
1.5 
0.0 
0.6 
0.0 
3441 
Mixed 
Decid./ 
Conif. 
10.4 
6.5 
2.0 
69.1 
11.5 
0.0 
0.5 
2217 
Deciduous 
1.5 
0.0 
0.0 
11.9 
85.9 
0.1 
0.6 
2899 
Clearings/ 
Cultures 
2.2 
0.0 
0.0 
2.7 
10.3 
74.5 
10.3 
224 
Tab. 5: Classification of control and 
training areas for six forest 
classes within the Regensburg 
map sheet, results in percent 
of verified area. 
independent control areas above. The 
comparison of confusion matrixes showed 
that five of six classes gave similar 
accuracy values, differing no more than 
4 %. The evaluation for the class of 
mixed spruce and pine provided a value 
of 60 % for Nuremberg and of 63 % for 
Augsburg. 
The classification accuracies for the 
Regensburg map sheet can be noticed in 
Tab. 6, showing the results for the 
combined verification and training 
areas. In this study site, the class of 
mixed spruce and pine shows a stronger 
overlapping with spruce stands. For the 
edition of the forest map of Regensburg, 
the classes of spruce and mixed spruce 
and pine were combined, which delivered 
a classification accuracy of about 88 % 
for this class (Keil et al., 1990). 
5 CONCLUSION 
Forest mapping with satellite data 
within Bavaria has delivered practical 
knowledge on large-area mapping. 
For small areas, high accuracies around 
90 % can be achieved with 
classifications according to different 
types of stands, to some extent also 
according to tree age class (Schardt, 
1988) . 
Less accuracy in classification can be 
expected for large-area mapping since 
spacial changes in the spectral 
signatures are registered much stronger 
here. An average classification accuracy 
of 81 % was calculated for the five or 
six respectively represented forest 
classes in the map sheets of Regensburg, 
Nuremberg and Augsburg. 
The large area mapping on the three map 
sheets showed that the ability to trans 
fer training area statistics onto large 
areas is coupled with several 
constraints. Thus, changes in the 
reflection values of forest stands due 
to varying atmospheric conditions must 
be taken into consideration. One must 
also consider the influence of the 
different geological site conditions on 
the reflection. Therefore it is 
important to register the varying 
spectral signatures in the forest as 
representative as possible. It may be 
necessary for the classification to 
divide up the entire area being studied. 
An increase in classification accuracy 
could be gained by integrating 
additional information. Thus a 
considerable improvement in the separa 
tion of pine and spruce could be 
achieved by dividing up the study areas 
according to the growth district map. 
Also, other additional information such 
as digitized site maps or, especially in 
mountainous areas, digital terrain mo 
dels can help towards an improved 
classification. One must always keep in 
mind whether or not the improved 
classification result is worth the 
increased amount of work involved.