each meant
ation classes
of which can
ed by STATS.
rithmetical
| object iden-
overlaps with
ated first
e masked i.e.
cation.
t was pre-
ind use cate-
ral signatures
ived from the
neous training
tial that
ed on topo-
mbers of the
| longitude.
miting the
1ly deduced
> the same
ita were then
s) and a map
iin landmarks
r agglomera-
nized.
intually be
te pixels can
tricted to the
responds
X £ pixels, map-
> best con-
pixels large
1s relatively
(ls within
ing often
] then be
4.2. Selection of training areas and location of training pixels
Training areas for the establishment of the spectral signatures had to
meet the following requirements: they had to be sufficiently large, easy
to locate and their radiance had to be relatively homogeneous over the
entire surface and in all spectral bands.
The minimum surface was set empirically at 10-12 pixels i.e. 4 to 5 ha.
Moreover the form of such small areas had to be more or less square in
shape in order to reduce contamination by border effects due to pixels
straddling limits.
Spectral homogeneity was checked after location on NMAPW prints of the four
spectral bands. This location was achieved by interpolation between recog-
nized reference points ‘taking into account differences in scale and print-
out deformations. This was greatly facilitated by the use of a B & L
Zoom Transfer Scope which allows for anamorphic correction. Discrepancies
between the LANDSAT data and the 1966 topographic map were checked in the
field, always to the advantage of the former.
4.3. Ascertainment of spectral signatures
The data corresponding to the pixels of each training area were then
entered in the STATS programme. With the radiance averages calculated
it was possible to establish the radiance range for each category in each
spectral band. Results are illustrated in fig. 2 and 3.
These radiance ranges are the main input for the EXIO programme. The other
statistical data derived by STATS were used as inputs for the CANAL programme.
4.4. Assessment of land use distribution
As radiance overtaps for the four categories retained, both on the March
and the May scene, are rather small in band 5, its information was used as
sole input for EXIO version I (fig. 5 and 7).
The information on band 7 for the May scene was treated slightly different-
lv by EXIO version 2. In this band there was indeed an interference of
cultivated land with the other categories. The cultivated land pixels.
well separable on band 5, were therefore eliminated prior to the separa-
tion in band 7 of the other categories (fig.8).
The CANAL programme was applied both to the March and to the May scenes.
The main purpose was to compare the results obtained by a supervised
classification programme makine use of the information in all four spec-
tral bands with one using onlv the information that appeared as having
the greatest diagnostic value (fig.6 and 9).