the whole of Hong Kong. This has not only increased the processing cost, but 
also created some difficulties in matching two spectrally different scenes. 
Fortunately, the Landsat scene E-21380-01562 acquired on 2nd November, 1978 
did cover the whole of Hong Kong (Fig. 1). Although there were a few 
scattered clouds aligned in a NE-SW direction across the area, it was decided 
that these were not detrimental for employment in land use mapping at 
1:100 000 scale (to cover the whole of Hong Kong and its New Territories) and 
at 1:25 000 scale (to cover just the major metropolitan areas of Hong Kong and 
Kowloon). 
In addition, another Landsat CCT of the same area (E-1520-02210) 
acquired on 25th December, 1973 was employed for analysis. The quality of the 
data in this case is excellent, but about one-quarter of Hong Kong to the east 
is not covered by this scene. This scene was employed only to produce the 
urban land use map at 1:25 000 scale in order to test the repeatability of 
land use mapping by the computer-assisted method and to examine the effect of 
temporal resolution on the accuracy of the resultant map. 
LAND USE CLASSIFICATION SCHEME 
The accuracy of a land use map should be evaluated with reference to the 
land use classification scheme adopted for mapping. This should further be 
judged in relation to the level of generalization intended and the scale at 
which the map is to be produced. For Hong Kong, there are only three 
official land use maps for the whole of Hong Kong published in 1958, 1966 and 
1977. Of these, the 1966 map with 24 categories of land use is the most 
detailed of all, and clearly exhibits Level II degree generalization of the 
USGS scheme (Anderson et al., 1976). In designing a suitable land use 
classification scheme for computer-assited mapping of land use for Hong Kong, 
the spatial resolution of 76 m (which is equivalent to 0.578 ha at the nominal 
altitude of 918.6 km) as given by Slater (1980) has been taken into account. 
It is noteworthy that the minimum size of mapping unit in the 1:250 000 and 
1:100 000 land use/land cover maps of the U.S.A. is 4 ha for all urban areas, 
bodies of water, surface mines, quarries, gravel pits and certain agricultural 
areas; but it is 16 ha for all other categories (Place, 1977). Therefore, the 
spatial resolution of the Landsat MSS data should be able to meet this 
minimum size of mapping unit, and some amalgamation of pixels together is 
expected if a less fragmentary map is to be produced. For the present 
application to Hong Kong, a nine-category land use classification scheme 
whichismodified after that for 1966 map is adopted for mapping at 1:100 000 
scale (Table 1). Another scheme of eight categories is also devised 
specifically for mapping the land use of the urban area at the scale of 
1: 25 000 (Table 2). The main purpose of these scheme is to bring out the 
essential land use patterns of Hong Kong. 
COMPUTER PROCESSING PROCEDURES 
The processing of the Landsat data was performed with the aid of the 
programs developed by the Office for Remote Sensing of Earth Resources (ORSER) 
of the Pennsylvania State University specifically for the analysis of 
satellite and aircraft multispectral scanner digital data (McMurtry et ai., 
1978). These programs were implemented and adapted for use with the IBM 3031 
computer of Hong Kong Universities and Polytechnic Computer Centre. The final 
products are a series of line-printer maps which can be quite cheaply produced. 
In order to investigate the possibilities of improving the accuracy of 
the resultant land use maps, some data preprocessing has been carried out. It 
should be noted that the Landsat CCT data from the multispectral sensors have 
already been radiometrically and geometrically corrected (U.S. Geological 
Survey, 1979). But the additional data-preprocessing carried out is to 
910 
M A 
—— ÁÜ—— r— Z 4 
= ee 
4