‚and sat 
cate- 
an area 
con- 
|. in the 
re a 
Ssses 
rban 
in the 
be easy 
he 
hat 
ctral 
to a 
ners 
e sense 
r hand, 
are 
inge 
area. 
are 
«Tt 
  
units and hence more mixed pixels. 
The computer cannot separate between urban fringe pixels and rural 
pixels, but it is possible to eliminate manually the rural pixels 
to see how much urban land is still included in the fringe pixels. 
If we assume the urban area defined by USEMAP as correct, the 
comparable Landsat urban area (excluding parks, sports complexes 
and detached houses under complete tree cover) should contain 
1962 ha. 
The delineated area of Landsat is without fringe pixels 1735 ha, 
with fringe pixels 2059 ha. The fringe pixels have therefore an 
acreage of 324 ha, of which 127 ha or 40% is 'rural'. This figure 
substantiates the hypothesis that the fringe pixels are mixed pixels 
at the edges of the urban area. 
The problem here is that in operational applications, manual elimi- 
nation of 'rural' pixels in this class would be highly impractical, 
so that the surface of the urban area will be either overestimated 
(if pixels around the separation line are included) or underestimated 
(if all pixels around the separation line are excluded). 
6. Conclusions 
By digital classification it is possible to obtain from LANDSAT a 
visually acceptable delineation of the urban area. Such delineations 
may be useful for specific applications, like the updating of small- 
scale maps, the monitoring of physical urban growth at national and 
regional scales and the definition of census tracts. 
Quantification of the urbanized area remains problematic and may 
contain considerable under- or overestimates, due to fringe pixels 
and differences with conventional definitions. Consistent subdivision 
into usable classes may be expected only for particular types of 
cities, but it is unlikely that this will be applicable nationwide 
or even regionwide, as smaller cities tend to have smaller land-use