ferous stands, Apparently, these stands comprised a maximum of
10% of the mixture, but really their percentage must have been
higher as is visible in the computer classification.
It was not surprising that class 1, the pinus silvestris forests
were underestimated. The pinus silvestris area within the test
area is relatively small, long stretched and surrounded by de-
ciduous stands or agricultural land. It follows that a relatively
large number of pixels along the border of the Pinus-stands are
not "clean", but highlighted by the neighbouring deciduous stands
or the various agricultural fields. This mays exolain the mis-
classification of 7 resp. 13$ of forest type 1.
3.4.2. Unsupervised Minimum Distance Classification
In an attempt to estimate the average precision of the unsuper-
vised minimum distance classification the same procedure was
used as previously. The test area was also the same as in the
test before, According to the line printer sheet 13 classes were
classified both forest and non=forest. For the purpose of this
study only four of these classes were considered, namelv the
same as used in the supervised classification.
The results of the unsupervised MD-Classification are shown in
table 7 and should be compared with the data in table 6.
Table 7 Confusion table used for test area 1 = Unsupervised
Classification
Forest Test Test Squares Classified Correct Omission
Type Squares as Type Classified
n 1 2 3 (4) n=% n=%
pinus silv. 100 70 15 0 15 70 30
deciduous 100 0 87 0 13 87 13
mixed conifers 100 0° 0 87 13 87 13
unclassified 0 0 0 0 0 : Le
Comission
n=% 0. 15 0 41
Inventory
Result n 300 70 102 87 41
Deviation of
n -30 .*2 4-13 *41
Average accuracy performance 81,33
It can be clearly seen that the difference in the average accus
racy performance between supervised'and unsupervised methods for
this test area is relatively small. Nevertheless, one can find
the better result in the supervised classification. From the
forest services' point of view omissions or commissionsof 10 to
20% - as found with the supervised classification - may be,