Abstract
For reducing the labor cost involved in vegetation mapping, we attempted to identify the forest types in forest areas by using high-resolution IKONOS satellite data. Thus we conducted a vegetation survey of 96 plots in Satoyama forest area and classified the plots into 8 forest types on the basis of the dominant tree species in the canopy layer. Cryptomeria japonica, Pinus spp., Castanopsis sieboldii, Quercus myrsinfolia, Q. serrata, Mallotusjaponicus, Phyllostachys bambusoides, and Phyllostachys heterocycla were detected as the dominant species representing the 8 forest types. The size of a plot was 16 m × 16 m and corresponded to 16 pixels of the IKONOS data. We developed classification tree models for each forest type by using 5 explanatory variables as follows: Normalized Difference Vegetation Index (NDVI), mean of the red band, standard deviations of the near-infrared band, red band and green band. Each model showed a good performance (area underthe curve, [AUC] > 0.8), except that by M. japonicus type. Each model involved a characteristic combination of the explanatory variables. The standard deviations of the near-infrared band and red band were adopted as the explanatory variables in some models such as that for Phyllostachys bambusoides type. Thus, textural property was also shown as useful for the identification of forest types when using the high-resolution satellite data; 72% of the plots were identified correctly. This result indicates that high-resolution satellite data could be used for mapping the Satoyama forest area.
