Figure 3. Road map updated from satellite data. Same part of study area as Figure 1, 2 and 3 (4.4 x 2.6 km). 
A = New roads; Present in old map: B = Arterial roads, C = Distributor roads, D = Neighbourhood roads, E = Railway. 
Separation of high density residential areas from vegetation was 
considered difficult during the interpretation, although this is not 
evident from the accuracy assessment. The grey shades are 
similar, and the structure evident in the image for the less dense 
classes is not pronounced in the high density areas. The road 
map was valuable to help in this delineation. 
The labelling of the residential areas had to be guided by field 
work in the study. It is expected that aerial photos and old land- 
use maps would have been very useful for this task, had they 
been available. It was shown that delineation of land-cover zones 
is possible in satellite images even with little reference data and 
familiarity with the area. When this ancillary information is 
added in the analysis - as would be the case in operational 
applications - better results are expected, both regarding accuracy 
and the number of classes possible to separate. 
The multispectral image was, as expected, better for identifying 
and delineating non-urban land-cover types. It was also better for 
separating high density areas from vegetation due to the 
contrasting colour shades. As it was judged to be considerably 
less useful for the urban land-cover classes, the panchromatic 
data would be the best single data source for this type of 
mapping. 
356 
Revised Road Map 
Field-checking of the new roads in the revised map at.1:25 000 
scale (Figure 3) gave an accuracy of 87% according to road 
length (Table 3). The errors of commission are most common. 
These include walls built around estates (40%), temporary tracks, 
and small drainage canals. New roads that were detected but had 
mistakes in the delineation are listed separately in the table. 
  
  
  
Length (m) % 
Correct 34760 87 
Delineation wrong 710 2 
Commission error 3880 10 
Omission error 730 2 
Total 40080 
  
Table 3. Field-check of new roads. 
The results must be regarded as good, considering that the 
interpretation was done without any reference data or knowledge 
of the area. The accuracy was comparable to that of the existing 
1:25 000 scale base-maps that had been produced from aerial 
photography with limited field checks. It was better than an "up- 
to-date” road map that was being offered to the general public in 
book-shops.