3.4, Digital Computer Aided Classification
The computer aided classification has been done with the DIBIAS
of the DFVLR in Oberpfaffenhofen. The authors gratefully acknow-
ledge the cooperation with the DFVLR and especially with
Dr,.Kritikos.
Only single date data (band 4, 5, 7) were used.
3.4.1. Supervised Maximum Likelihood Classification
The test area was located in the Rhine Vallev and its foothills.
Definite training areas were selected for three main forest types.
The classified units are shown in table 5, odumn ^ «AD
attempt to classify more,and more detailed forest types was
without significant success.
The accuracy of the obtained results were tested by two different
methods. The first used planimetering of the ground truth maps
and then compared the results with those of the maximum likeli-
hood classification obtained from the line printer sheet. The
results of the acreage determination and comparisons are shown
in table 5.
Considering the ground truth as the starting point for estimat-
ing the computer's mapping accuracy and assuming that each class
unit corresponds to an accuracy of 100$, the corresponding class
obtained via computer will have an accuracy of X. From this con-
cept, 13.760,00 ha, of the forest land in the base-mao corres-
pond to 100%. Since 15.272,00 ha. were obtained by the automatic
process, there exists a commission error - that is, more forest
resources were interpreted than the existing ones. The mathema-
tical relation is an inverse one, which enables us to obtain an
overall accuracy performance of 89,00$. Considering the same
mathematical relation for the different classes, an accuracy
was obtained for mixed coniferous of 93,40$, for mixed-deciduous
of 83,209 and for P.silvestris of 86,502. The average class per-
formance was 87,179 (about 1,30$ smaller than the overall accu-
racy).
Table 5 Comparison of acreage determination between ground
truth and computer classification
Forest Ground Truth Computer Error
Type Classification
ha. $ 3 Pixel ha, % $ $
Mixed
Conifers (mount.) 6215 45,2 13,5 14620 6624 43,4 14,4 +6,6
Mixed
Deciduous 7013 51,0 15,2 18072 8188 53,6 17,8 +16,8
EE 532... 3,9," 1,2 1015... 460 3,0 1,0 -1$,5 i
Forest Land total 3760 100,0 29,9 33707 15272 00,0 33,2 +11,0
Unclassified area 32240 70,1 67821 30728 66,8 — 4,7
Test area total . 46000 100,0 101528 46000 100,0
