her
si
ile
;rial/
ty.
> both
ceta-
vphy
ninor
irison
amme s
Tber-
irther
each
sually
50%
3 may
le-
again.
Finally, & field check has been carried out. Results of the interpre-
tation are shown in Fig. 1. Only the main land-use classes have been
mapped here.
The USEMAP interpretation was intended and produced as a full,
highly detailed land-use interpretation. Thanks to this detailed
interpretation, it was possible to aggregate the 65 original classes
into 2 classes urban/non-urban that coincide closely with the urban/
non-urban classes that digital processing could provide for Landsat.
4. Delineation of the urban area from Landsat
The method used to define the urban area has been application of
principal component analysis. This method has been described in
MULDER (1975) and in DONKER and MULDER (1977). The various steps in
the method are indicated in Fig. 2.
The information of the original 4 bands of the Landsat frame taken
on 22.03.73 is reduced to 2 principal component (pc) maps containing
nearly all the original information. Each pixel may now be plotted
in à 2-dimensional 'feature space diagram! according to its values for
pcl and pc2.
If representative pixels are sampled for selected land uses, they may
form separate clusters in the feature space diagram. If a cluster is
not mixed with other clusters, it means that the land uses can be
easily separated spectrally.
Fig. 3 gives the feature space diagram for the Enschede area. The
clusters forming the urban area (residential, industrial and water)
are overlapping; the clusters for forest and agriculture are separate.
It is now possible to construct a 'separation line' (S4 in Fig. 3)
that separates the urban land uses from the rural land uses.
The following step is to produce a thematic map of the urban area
by printing black all the pixels that are located in the feature
space above the separation line and blank all the pixels below the
line.