For the ground truth preparation, field control data, Color
Infrared Airphotos (scale 1:11,000) as well as scanner infrared
black and white images (several scales) were used, The ground
truth was transfered to a 1:25.000 base-map by means of an
Aerosketchmaster. The acreage was made through a polar plani-
meter.
Four methods of interpretation were tested, two of which were
qualitative ones: a visual conventional interpretation by using
the traditional patterns of aerial photointerpretation and an
interpretation through a Color Additive Viewer I2S model 6000.
And two quantitative ones; the first by means of a Joyce-Loebl
microdensitometer model 3 CS. The second, by unsupervised as
well as supervised computer-aided classification methods.
The tests were based on Landsat-MSS data scene 2199-09374, bands
4, 5, 6 or 7 respectively, from August 9, 1975. According to the
various investigations black and white bulk images 1:200.000,
transparencies 1:800.000 or CCT's have been used.
3. Results. and Discussion:
3.1. Visual Conventional Interpretation using black and white
bulk images
The process of interpretation followed the principle of a dicho-
tomous key and is illustrated in figure 1. In this process mainly
bands 5 and 7, taken in summer were used. These bands were found
to be the most informative ones for identifying forest types in
the area.
Following this key, it was possible by visual interpretation to
discriminate forests from non-forests, to separate deciduous
forests, mixed forests, pine forests, spruce-fir and/or pure
spruce forests.
The identification of tree species was only possible for Pinus
silvestris when growing in large pure stands and for Larix Sp.
in forests with large portions of these Species. The latter,
when using band 7 images taken in fall in addition to the summer
images.
A quantitative analysis was made, based on 31 sample units, each
one measuring 400 ha. For this analysis the area was stratified
into the conifer-, the mixed forest-, and the deciduous forest
region. Based on a previous pilot sampling of the forest population,
the optimal number of units for each region proportional to the
whole population was estimated and the units within the regions
were randomly allocated.
For each of the 31 units three type maps were produced,
one based on the ground truth (CIR airphotos at a 1:11.000 scale)
the second and third based on the interpretation of Landsat
images from bands 5 and 7 resp. taken in summer.