Vegetation mapping using airborne hyperspectral sensor data measured in early summer

Abstract

The authors previously made a vegetation map that included beech, walnut, dwarf bamboo and altherbosa, using airborne hyperspectral sensor (HSS) data acquired in early autumn in the Shirakami Mountains of Japan. From the spectral response pattern, high and low radiances were identified for beech and for walnut, so we created four classifications for these: high-radiance beech, high-radiance walnut, low-radiance beech, and low-radiance walnut. However, field observations found no visual difference (e.g., tree age) between the high-radiance and lowradiance areas for each of these vegetations. The difference in radiance may depend on differences in the photosynthetic capacity of foliage immediately before the foliage changes in color in autumn. In the present study, we made a vegetation map for the same area using airborne HSS data from early summer. From the map, differences in radiance were not found in beech vegetation and walnut vegetation. The producer's classification accuracy of walnut increased slightly to 78.5% in early summer, from 76.5% in early autumn, and the producer's classification accuracy of beech increased to 96.9% in early summer, from 80.0% in early autumn. In addition to the improvement of classification accuracy, the shade area in the tree crowns is smaller in summer than in autumn, and it is easier to extract training pixels from airborne HSS data of tree crowns in early summer than in early autumn. Judging from these facts, it was found that airborne vegetation classification maps drawn from HSS data are more accurate when the data are acquired in summer than when the data are acquired in early autumn.