An experimental study on the extraction of the yearly variation of land surface was made using both of Landsat data and the land use grid square data of Land National Digital Information. The test site was the West of Tama Kyuryo and the data used were Landsat MSS data taken in 1979 and the land use grid square data KS-202.
As the algorithm for the classification of Landsat data, the following two kinds of method were tried, Methodl in which the training data sets are generated automatically and Method-2 in which they are assigned by human interpretation. Using the Landsat data classification results and the land use data, five kinds of variation patterns of land use were extracted in the test site. The result of the extraction was evaluated by the ground survey, which resulted in a fairly good correspondence.
Among the classification algorithms, the Method-2 using Exclidean minimum distance resulted in the best result of the extraction, however, it was a remarkable result that the Method-1 also resulted in the approximately same result of the extraction as the Method-2 in spite of its automatism and low discrimination score of image data.
View full abstract