Vegetation was classified using aerial hyperspectral sensor (CASI-3) data and field survey data for the beech-forested Shirakami Mountains in Aomori Prefecture, Japan. Previous study on German forest reported that different plants show similar hyperspectral response patterns and that the accuracy of vegetation classification using aerial hyperspectral sensor data alone was 66%. In this study, radiance in each grid was divided by average radiance in a vegetation-coverage area. This“normalized”spectral response pattern enabled finer plant discrimination than the non-normalized patterns used in the previous study.
Plants typifying those in the study area were selected, their radiances were investigated at ground truth sites, and eight vegetation classifications were selected : walnut (high and low radiance), beech (high and low radiance), altherbosa, dwarf bamboo, bare ground, and other. Classification accuracy was 77.6% at the ground truth sites, but the classification accuracy for each of altherbosa, bamboo grass, and bare ground was 100%. Walnut classification accuracies (63.6% for high radiance and 50.0% for low radiance) were equal to or lower than beech classification accuracies (50.0% for high radiance and 88.9% for low radiance) . When radiance is disregarded, classification accuracy was 76.5% for walnut and 80.0% for beech. The classification accuracy of 76.5% for walnut 80.0% for beech suggests that aerial hyperspectral sensor data can be used to map vegetation.
