比演算を施した2季節のランドサットMSSデータによる広域山地の林相区分

抄録

This paper refers the feasibility of Landsat MSS data for forest type classification. Study area was chosen at Tango penisula of central Honshu island in Japan. Forest type classification was performed using two season combined Landsat MSS data after the ratioing process between two adjacent spectral bands. The reasons to use this sort of data are as follows ; 1) Value of Landsat MSS image varies with the ground slope and direction. To eliminate this slope effect in the data analysis, the ratioed value between two adjacent bands can be applicable. 2) Spectral reflectance of leaf varies with season. Combination of two seasonal Landsat MSS data before and after autumn tints will cause a distinct separation between an ever green leaf and a deciduos one. Requirements for the two-season combined Landsat MSS data are as follows ; 1) Line noise in the raw Landsat MSS data seen like as "stripe on the cloth" due to 6 detectors, should be reduced as low as possible. Radiometrically uniform MSS data within about 0. 2 in quantum level were obtained by the level matching method using cumulative frequency distribution. 2) There needs rigorous registration between two seasonal Landsat imageries. Using 15 ground control points around the study area, a registration accuracy with about one pixel, i.e., 57 meters in standard deviation was obtained in the two Landsat imageries and the corresponding 1:50, 000 topographical maps.
Comparison of the classified results with the forest statistics gives the following facts; Classification errors of each forest type became about 7% in Conifer to 13% in Hardwood in average for each local government area.