Abstract
This paper attempts to extract damaged areas due to two forest fires occurred on April 27 in 1983 simultaneously in Tohoku District. The test sites are Kuji-shi in Iwate-Ken and the northen part of Sendai-shi in Miyagi-ken, which includes Sendai-shi, Izumi-shi, Rifu-cho, Tomiya-cho and Taiwa-cho.
The supervised maximum likelihood classification method is employed to extract forest type and land cover distribution from the MSS data taken before the forest fire and to extract the damaged area from the data taken after the forest fire. By combining the two classification images, the areas for forest damage are computed by the degree of damage and by forest type. Also, the damaged areas by administrative district are computed using the administrative district mesh data of Digital National Land Information. As an alternative method of supervised classification, an unsupervised classification, cluster analysis, is tried to estimate the damaged forest area.
These estimated areas by Landsat data are compared with the statistical report by Miyagi and Iwate Authorities. As the result of comparison, about plus/minus ten percent accuracy was obtained in total damaged area estimation at the test sites using both of supervised and unsupervised classification method, although the error includes both of the error for forest area estimation and the damaged area estimation.
The estimation error is modified using the forest area estimation error obtained by the comparison between Landsat classification and the land use data of Digital National Land Information. According to the result of this modification, very accurate estimation which error is lower than one percent is expected in total damaged area estimation using the supervised classification method. In the case of unsupervised cluster analysis, the estimation accuracy is considered to be rather lower than that of the supervised classification, however, the method is expected to be useful in the application of Landsat data for forest disaster assesment in foreign countries.
