Abstract
As a part of the application of remote sensing technique on grassland management, Landsat Thematic Mapper (TM) data over the Nishinasuno township was analyzed using the principal component analysis for the image processing and land-cover classification. The first principal component score mainly represents the features of bands 1, 2 and 3 (visible), that is, brightness. The second score mainly represents the features of band 4 (near-infrared), that is, greenness. And the third score mainly represents the features of bands 5 and 7 (mid-infrared). By use of the first two or three components, it could involve 91.5 or 98.6% of the total information contained in six TM bands. False color image derived from the first three scores was better than that from the original three bands, because it was easier to discriminate between grassland and forest, between grassland and bare soil, and between meadows before and after mowing. The results of land-cover classification by maximum likelihood method indicated that the principal component scores were better for classification than original bands under the condition that the same amount of data were used. Reduction of variate dimensions by the use of principal component resulted in the merit that analyzers could visually comprehend the classification area.
