mapping. Examples include Thibault ez al. (1991) who 
delineated 15 urban land-cover types; and Gamba (1990) who 
created a typology comprising 38 zones based on structure, 
density, land-use, and vegetation type. 
For operational applications, visual interpretation would therefore 
be the obvious choice. This paper examines the utility of SPOT 
data interpreted with limited access to reference data. 
URBAN ROAD MAPS 
As for urban land-cover mapping, methodologies for automated 
road delineation from high-resolution satellite data have started to 
be developed (Van Cleynenbreugel et al., 1990; Moller-Jensen, 
1990), but the results do not yet match those achieved by visual 
interpretation. 
Through visual techniques, major roads and railways are readily 
detectable, while minor roads and tracks cannot be reliably 
mapped (Salgé and Roos-Josserand, 1988; Manning and Evans, 
1988). Primary reasons for non-detection are tree canopy closure 
or weak contrast between surface material and the surrounding. 
Alleys in very high density urban areas, which are common in 
developing country cities, are also difficult to distinguish 
(Bertaud, 1989). 
The higher contrast of recently constructed roads suggests that 
satellite imagery would be valuable for revision of urban road 
maps. A test of this application is reported here. 
STUDY AREA AND DATA SOURCES 
An area of rather flat terrain covering 8 by 8 km on the southern 
edge of Metropolitan Manila, Philippines, was selected for the 
study. This displays a variety of land-uses, and contains both 
older areas and significant recent growth in urban development 
(Figure 1). 
Three digital SPOT images were used in the study; two 
panchromatic from 21 February 1987 and 12 February 1990, and 
one multispectral dated 18 February 1989. The 1990 image was 
geometrically corrected by resampling to UTM projection on 1:25 
000 urban base-maps. The RMS error was less than one pixel. 
The other two images were resampled to the corrected image 
with an error of less than one half of a pixel. 
The base-maps at 1:25 000 scale were used as reference data in 
the analysis. They were produced from aerial photos dated 1986 
and contained information on roads, rivers and administrative 
boundaries. For evaluation of the results, land-use maps at 1:10 
000 scale were used. These were produced from aerial photos 
dated 1983 and 1986, but were published only in 1989/90. 
Although of good quality, they had to be supplemented by field 
work to incorporate changes, and are therefore a good example 
of urban maps that are outdated already when published. 
353 
IMAGE ANALYSIS 
The panchromatic image from 1990 was used for the land-cover 
mapping. With only one analyst, it was not considered possible 
to do an unbiased interpretation of both panchromatic, 
multispectral, and multidate imagery for comparison of the 
information content in the data. Instead, the most promising data 
were chosen for the mapping. The multispectral image was 
compared to the panchromatic after the interpretation was 
completed to decide relative advantages. 
The digital imagery was interpreted on the monitor in an 
integrated image processing/GIS system. The digitized road 
information in the base-map from 1986 was overlaid on the 
imagery to help the analysis. While the scale of the map was to 
be 1:25 000, interpretation was done at 1:12 000 and 1:6000 
scale. 
Areas judged to be homogeneous regarding land-cover were 
delineated and interactively digitized. As little reference data was 
available, the polygons were only given tentative labels for 
checking in the subsequent field work. The interpretation of 
land-cover relied mainly on the size and layout of buildings, 
image texture, context, and the structure of the area. Special 
emphasis was put on separating types of residential areas and 
dynamic classes describing the on-going developments. The 
latter would be important to enhance and maintain the usefulness 
of this type of map in a fast developing area. 
For updating of roads, all the three SPOT images were used 
sequentially. The road network according to the digitized base- 
map from 1986 was superimposed on each image separately, and 
new roads delineated and digitized directly on the monitor in the 
combined image processing/GIS system. 
ACCURACY ASSESSMENT 
The accuracy of the land-cover mapping was assessed by 
comparing it to the reference land-use map at 1:10 000 scale. A 
regular grid was used to select the sample points. First, the 
classes in the reference map had to be converted to agree with the 
nomenclature in the land-cover map. As the sources for the 
reference map were from 1983 and 1986, the code for areas that 
had changed were adjusted according to field work findings. 
Densities of the residential areas were derived from the building 
outlines present in the reference map. 
The accuracy of the road map had to be checked through field 
work. Around 60% of the roads delineated from the satellite 
images were visited. Roads missing in the map were recorded 
when encountered. A systematic search for missing roads was 
not feasible for the whole area, but it can be assumed that most 
new roads would be close to areas of new development, and 
therefore that most missing roads were found in the parts covered 
by the field visit.