リモートセンシングデータの空間分解能と住宅地の土地被覆分類結果との関係 : ミクセルの観点からの考察

Abstract

We have assigned the components of urban residential areas to the land cover categories, and discussed the actual conditions of MIXELs and the effect on classification accuracy selecting two types of residential areas. This paper first, clarified the relation between the ratios of MIXELs and spatial resolution using a land cover grid image. The results are as follows: (1) The ratio of MIXELs are 20-40 % at resolution of 1 m. This distribution follows a increase according with the coarsening in spatial resolution. In an image with less than 5 m of resolution, there is little PURE PIX-EL and almost all MIXEL. (2) We divided MIXELs into two groups: MIXEL-A which are MIXELs not containing detached parcels, MIXEL-B which are MIXELs containing detached parcels. At less than 2. 5 m, most MIXELs are MIXEL-A. At 5 m some amount of MIXEL-A changed into MIXEL-B. The ratios of MIXELs show little alteration between 10m and 30m resolution, but MIXEL-B became predominant at 10m in the DETACHED HOUSING AREA and, in the HOUSING COMPLEX AREA, MIXEL-A is still predominant at 30 m. Second, using simulation images with various resolutions the ratio of pixels classified in the different class category from a MSS image with the original 1. 25 m resolution was obtained. The ratio is inversely proportional to spatial resolution and reduced by half for a 2. 5 m resolution. Finally, the effect of MIXELs on land cover classification results was assessed. As a consequence at less than changing point resolution the increase in the ratio of MIXEL-A caused a decrease in the land cover classification accuracy and at above the resolution that of MIXEL-B did.