Application of automatic binarization method for nationwide forest area mapping using satellite imagery

Article 3.3 of the Kyoto protocol obligates nations to report reduced amount of carbon dioxide (CO_2) emission, and allows including CO_2 absorption by Afforestation (A), Reforestation (R), and Deforestation (D) activities to the amount of carbon accounting. So it is necessary to clarify ARD location after the end of 1989. The primary objective of this study is developing a forest area mapping method using satellite images. We applied two classification methods, Decision Tree Classification (DTC) and Spectral Shape Classification (SSC) to classify forest area using a Landsat Thematic Mapper image over a test site in Higashi-Shirakawa village in Gifu Prefecture. An automatic binarization of digital number of a band (band 3 or band 7) and the normalized difference vegetation index was applied in DTC. The accuracy was evaluated using a ground truth map which was produced using aerial orthophotos by a visual interpretation, and DTC gave a slightly more accurate result than that of SSC. The binalization using band 3 showed better classification between forest and non-forest area than that using band 7. Moreover, DTC was better than SSC in classifying forest and non-forest area using two Enhanced Thematic Mapper Plus images in a validation of classification stability. Therefore, we concluded that DTC gave more stable results than SSC.