Abstract
Landsat multispectral scanner (MSS) data acquired on May 22 in 1979, over Northern Tochigi district (Figs.1 and 2) was analyzed using the Image Information Processing System at Chiba University to evaluate the utility of satellite data for land-use classification. A flow chart of the basic steps taken to perform this analysis is shown in Fig.3. In addition to the original four Landsat MSS bands (500-600nm, 600-700nm, 700-800nm and 800-1100nm), two rationing treatments using bands 5 (red) and band 7 (infrared) were applied in this analysis (Table 1). Fourteen cluster classes were identified as exhibiting spectrally separable characteristics and they merged finally into nine types of land-use groups. The number of cases classified into each cluster class and land-use group for training data is shown in Table 2. More than 90% performance in every group, and an average of 94.0% performonce for overall were attained. Difficulties were encountered in classifying groups such as "Forest"-"Grassland" or "Urban"-"Bare soil" correctly. These difficulties have currently been resolved by using Landsat data from different seasons. A land-use maps with six groups represented in different symbols covering the Nishinasuno Town was produced (Fig.4). The accuracy of the Landsat estimates were examined by contrasting them with the Real Estate Records in Municipal Offices of one city and two towns in Northern Tochigi Prefecture (Table 3). The general patterns of the classifications were found to be acceptable when we incorporate proper corrections for discrepancies which occurred because of differences between the present actual state of land-use and the registered land classification (Table 4).
