placed on the results. rer 
for 
Studies by FORSTER (1980, 1981) have shown that within the spectral range of 
the current Landsat sensors the basic urban surfaces are difficult to distin- 
guish because of overlap and the parallel nature of their spectral signatures, 
further increasing the difficulties of spectral separation. It would appear 
that in urban areas there are three distinct surface classes with sub-classes 
being differentiated by brightness alone. These three classes might be con- 
sidered as vegetation, urban residential (residential buildings and their 
curtilages) and urban non-residential (roads and other buildings) (FORSTER, 
1981, 122-123). 
An increase in the number of spectral bands available should aid surface pre- 
diction and differentiation, although the determination of the logical connec- 
tions between surface combinations and land use would still remain a problem, 
being primarily determined by spatial resolution (see point (a) previous). 
RESOLUTION REQUIREMENTS 
EVERETT & SIMONETT (1976) have suggested the resolution required to undertake 
certain urban monitoring functions,and apart from open space analysis, these all 
tend to require a resolution of 10 metres or less. Normal requirements for 
urban land use mapping from medium to large scale aerial photography suggest a 
ground resolution of less than 3 metres (WELCH, 1982) which would be sufficient 
to map class II (and possibly III) based on Anderson's four level classifi- 
cation system (ANDERSON et al, 1976). At a recent conference on primary data 
acquisition it was shown that a resolution of 5 metres or less was necessary 
for urban cartographic purposes (KONECNY et al, 1982). This would require a 
sensor system with an IFOV of approximately 2 metres. WELCH (1982) in a 
detailed paper, concludes, that an IFOV of 5 metres or better is most effective 
for visual interpretation and that truly significant gains in information 
extraction will occur only as IFOV's are reduced to less than 10 metres, 
particularly in non-western cities with higher spatial frequencies. 
In Sydney the average area of a residential parcel (allotment) is approximately FUTI 
750 m?, with about 5 houses per Landsat pixel (IFOV 80 metres). The density 
generally increases towards the CBD, although there are a number of more ex- Dur 
clusive inner areas with considerably larger parcels. An IFOV of 10 to 15 lau 
metres (say 3 or 4 pixels per parcel) would be required to map land use at the Sta! 
parcel level. D TI 
grot 
It may not always be necessary to achieve this order of resolution for thematic wil 
mapping (as compared to cartographic mapping) where detection may be more res 
important than identification. Primarily in urban areas the concern is with rest 
the road pattern as this determines the framework in which the complexities of ster 
urban land use change take place. The road elements make up cells considerably 
larger than the individual parcels. Within these cells there is a high The 
probability of a common land use. of r 
leas 
Using current Landsat data it has been shown (FORSTER, 1981) that housing 
density (over seventy 0.25 km? training areas) was highly correlated with It« 
Landsat response data with r = 0.91; other interesting relationships also with 
exist between average parcel size and the difference in the response from mixi 
Landsat bands 4 and 5. This is plotted in Figure 1 for seventy training areas, cert 
and apart from a few areas with small parcels and attached single family adva 
dwellings (up to parcel sizes of 30 m?), the relationship is relatively simple. 
These results suggest that it is not always necessary to have a small IFOV to (a) 
obtain information of a reasonably detailed nature and quite suitable for many 
purposes, for example, between census change. However it is still very desir- 
able that individual residential buildings can be detected because these 
$92