er-
were
is
in
ission
————
cue
pvossibly, negligible for certain practical purposes, but omis-
sions and commissions of 30 and 40$ - as found with the unsuper-
vised classification - are in any case insufficient.
Considering the different conditions in. forestry the results ob-
tained in chapter 3.4. (table 5, 6, 7) can be classified as si-
milar with those obtained by tvpical studies in North. America
(e.g. Kalensky 1974, 1976; Hoffer 1976; Mead and Mever 1977 etc.)
and Europe (Lapietra and Megier 1976; Jaakkola, 1976).
4, Conclusions
Considering the results obtained for the present study area,the
following conclusions can be taken:
1. Visual interpretation of black and white prints (1:200,000)
has not given so much information about forest typing, however
conifers and deciduous in bands 5 and 7 could be separated in
almost 100 percent of the cases.
Multi-date imagery could not be added in this investigation but
it became clear that the season of the year must be carefullv
selected.
2. Color composite interpretation besides being easier, added
more information in forest typing and its overall accuracy
(boundarv delineation) was similar to the first case.
3. Microdensitometer measurements were very useful in forest
typing (in very controlled conditions). Changes in crown closure
and age classes and in the overall structure of the different
stands, could be detected in the microdensograms with statisti-
cal significance.
4, The supervised method of automatic classification was more
accurate than the unsupervised besides being a little more
"operational". However, in areas similar to the present one,
where many small stands are cultivated, it is difficult to ex-
tract representative "clean" training sets statistics for
supporting the supervised approach. In this case, the unsuper-
vised classification oroved to be useful for obtaining an over-
all complete view of the entire area and its main forest tvpes.
5. It can be concluded that computer assisted forest mapping
using Landsat 1 + 2 data in areas like the present one is
possibly but not easilv to be accomplished. The results do not
meet all information requirements of an intensive forest
management system in terms of detail and accuracy. The results,
however, mav be useful for survevs at the reconnaissance or
pre-investment level,
Probably, a cluster modified procedure (Hoffer et.al. 1976) or
an interactive approach to eliminate obvious classification
mistakes could produce better results.