LAS
now-
11s.
ypes.
srent
ps
li-
Nn
at-
lass
lass
con-
5 —
atic
est
ma--
an
uous
per-
cu-
= 1804 -
As in the visual interpretation, the supervised classification
results in an overestimation of the total forest area. But
- compared with the visual interpretations - opposed (and not
sufficient) estimates have been obtained for the three "forest
types". Surprisingly the deciduous forests were overestimated and
the pine forest area, which could be clearly interpreted visually
in the color composite, was considerably underestimated.
Again one can conclude that, due to the available ground resolu-
tion too many pixels ru ain mixed information which do not meet
the spectral characteristics of the definite training areas of
the selected types. Especially the pixels which mainly represent
mixed forests have been preferably classified as deciduous forests
and also many agricultural areas were mis-classified as deciduous
forest.
Table 6 illustrates the results from the second method of mapping
accuracy estimation which used an overlay with a net of 1 x 1 cm
squares for both, the ground truth and the computer map. This
accuracy estimation results in a confusion table (table 6).
Table 6 Confusion table used for test area 1 = Supervised
classification
Forest Test Test Squares Classified Correct Omission
Tvpe Squares as Type Classified
n 1 2 3 (4) n=% n=%
pinus silv. 100 80 7 0 13 80 20
deciduous 100 2 89 0 9 89 11
mixed conifers 100 0 7 93 0 93 7
unclassified 9 0-8 0 0 - =
Commission E 2 14 0 22 - _
n=%
Inventory
result n 300 82 103 93 22 - -
Deviation n -18. 3 -7 +22 — -
Average accuracy performance in $ 87,33 -
Analyzing table 6 it can be observed that deciduous stands were
"confused" with the "class 4" (= unclassified areas) in 9% of
the cases. This is mainly due to the spectral similarity among
vineyards and grassland when compared with deciduous stands. This
was especially evident in areas with rugged topography.
With regard to mixed coniferous stands (class 3), the possibility
of interpretation seems to be better than for deciduous. Even so,
some "confusion" was also observed (7%), despite the distinct
spectral signatures of both covers. This problem could have
arisen during the ground truth preparation when some deciduous
stands (considered small ones) were introduced into the coni-