Utility of Very High Resolution Imagery for Forest Type Classification and Stand Structure Estimation

摘要

We compared the relative utilities of very high resolution imagery (VHRI) and medium resolution imagery (MRI) for forest type classification and stand structure estimation. We used QuickBird imagery for the VHRI with object-based classification and LAND SAT/ETM+ imagery for the MRI with pixel-based classification. The study site contained even-aged plantations of Japanese cedar (Cryptomeria japonica D. Don) and Japanese cypress (Chamaecyparis obtusa (Sieb. et Zucc.) Endl.) and natural broad-leaved forests. The overall accuracy of forest classification was 81% with the VHRI and 72% with the MRI; the VHRI was more accurate in discriminating Japanese cypress from natural broad-leaved forest. Stem density was not correlated with any features measurablefrom VHRI, whereas the texture measures had significant curvilinear relations with the stand volume of both Japanese cedar and Japanese cypress (the relative root mean square error was 8.6% and 18.0%, respectively). The pixel values of MRI were not correlated with either stem density or stand volume. We conclude that MRI is virtually not enough to use in forest management planning for practical use in Japan and the use of VHRI is recommended for it. Texture information is important for both classification and stand structure estimation to exert the potential of VHRI.