correspondence, Urban parks, however, tend to be smaller on Landsat 
than in reality, boundary pixels seem to fall into the urban cate- 
gory. USEMAP also has a much larger area for Stroinkslanden, an area 
east of the southern district. In 1972/1973, this area was a con- 
Struction site, delineated by the photo-interpreters too much in the 
anticipation of what was bound to happen, so Landsat gives here a 
better delineation. 
5.2.2 Subdivision of the urban area into meaningful classes 
Various attempts have been made to differentiate inside the urban 
areas different types of land uses. The overlapping clusters in the 
feature space (Fig. 3) indicated already that this would not be easy 
and results were indeed unconvincing. Partly this is due to the 
relatively small sizes of areas of homogeneous land use, so that 
there is a high percentage of mixed pixels!) partly also spectral 
differences in urban areas like Enschede will correlate only to a 
limited extent with conventional land-use classes as the planners 
need them. A class like 'medium dense urban area' makes little sense 
if it cannot be correlated consistently with terrain data like 
housing density, floor-space index, etc. 
It goes without saying that photo interpretation, on the other hand, 
will produce most of such data rather easily. 
5.2.3 Fringe pixels 
The pixels close to the separation line in the feature space are 
located at the urban/rural boundary in geographical space (fringe 
pixels), but they occur also at scattered places in the rural area. 
  
1) As an illustration, Fig. 7 shows which gridcells in USEMAP are 
homogeneous with regard to the 3-digit classification used. It 
can easily be seen that homogeneous cells occur mainly outside 
the urban area. 
In the case of a 1-digit classification (7 classes), there would 
be more homogeneous cells, but one would be surprised at the 
high number of mixed cells that remain. 
  
   
  
  
  
  
  
   
  
  
  
  
  
  
   
     
   
   
   
   
   
  
    
    
    
po as 
ns