404
reason, the verification results are
compared between results of independent
areas, on the one hand, and results of
combined control and training areas, on
the other hand.
Most of the independent control areas
cover homogeneous stand types, which
should be related to one class by 100
percent. The misclassification in these
areas can be shown in confusion
matrices.
Additionally, control areas were also
chosen for heterogeneous mixed
deciduous/coniferous stands. For these
heterogeneous verification areas, the
percentage of deciduous and coniferous
wood was estimated by the aerial photo
interpretation and compared with the sa
tellite classification. The resulting
differences were evaluated to a weighted
mean value which provides an estimate
for the classification accuracy in mixed
stands that are not homogeneous.
4 RESULTS OF CLASSIFICATION AND
VERIFICATION
In the three map sheets of Regensburg,
Nuremberg, and Augsburg the following
six forest stand types were derived:
spruce stands (also including the
subclass "spruce with snow and storm
damage"), mixed spruce and pine stands,
pine stands, deciduous forest, mixed
deciduous and coniferous forest, young
cultures and clearings.
Due to their poor separability, the
classes "spruce" and "mixed spruce and
pine" were later combined into one class
for the resultant Regensburg forest map.
The class "mixed spruce and pine",
however, was kept in the map sheets of
Nuremberg and Augsburg despite limited
classification accuracy in order to
preserve the information of their dis
tribution. The combined class "young
cultures and clearings" had to be
created for the various transition
states between presently unstocked areas
(e.g. clearings due to storm damage) and
not yet closed cultures.
The areal statistics of the above
mentioned six forest classes for the
three map sheets are given in Tab. 1.
The resulting percentages of the
classified stand types related to the
complete forest area are shown in
Tab. 2.
The verification results for the forest
classification in the Nuremberg and
Augsburg map sheet are presented in Tab.
3 and Tab. 4. The results were derived
on the basis of independent verification
areas.
Relating to the Nuremberg map sheet,
five of the six forest classes show
accuracies of more than 75 %; for
example, 86 % of the verified spruce
areas were classified accurately.
Forest Classes
Area in km 2
Regensburg
Nuremberg
Augsburg
Spruce
1253
602
1065
Mixed
Spruce/ Pine
472
333
86
Pine
564
658
54
Mixed
Decid./ Conifer.
684
587
565
Deciduous
239
417
516
Clearings/
Cultures
137
139
108
Nonforest
5195
5910
6389
Tab. 1: Areal statistics for forest
classification within the
three map sheets
Forest Classes
Percentage of forest area
Regensburg
Nuremberg
Augsburg
Spruce
37.4
22.0
44.5
Mixed
Spruce/ Pine
14.1
12.2
3.6
Pine
16.8
24.0
2.3
Mixed
Decid./ Conifer.
20.4
21.5
23.6
Deciduous
7.1
15.2
21.6
Clearings/
Cultures
4.1
5.1
4.5
Tab. 2: Percentages of classified
forest stand types within the
three map sheets
The verification areas representing
mixed spruce and pine stands were
classified accurately only by 50 %, a
high percentage of 32 % was classified
as pine stands. This demonstrates the
problems to separate mixed spruce and
pine stands in this area.
Two effects lead to the high amount of
misclassification between mixed spruce/
pine and pure stands of spruce an pine:
In the region around Nuremberg (growth
district Southern Keuper Declinity),
mixed stands of spruce and pine occur
mainly in two layered stands having
spruce of different stem heights in the
undergrowth. This effect lead to
variances in the grey values. Additional
superimpositions on signature of these
stands are caused by changing soil and
water supply conditions, influences of
age class and canopy density. On the
other hand, the low accuracy value is
partly caused by the strong heterogenity
of these stands which made it difficult
to mark off verification areas clearly.