EVALUATION OF THE POTENTIAL OF HIGH SPATIAL RESOLUTION SATELLITE IMAGERY FOR DETECTING OAK WILT INFECTED TREES

초록

Oak wilt is one of the serious forest diseases in Korea. In this study, we try to assess the potential of highresolution satellite imagery for detecting oak wilt damaged trees. Worldview-2 image acquired on May 28, 2018 was used for the study site in Incheon. Three datasets of panchromatic image, multispectral image, and, pan-sharpened image were separately analysed to detect the damaged trees. Several spectral indices were derived from the multispectral images, and they were used to detect oak wilt damaged trees by a simple threshold, spectral feature matching, and machine learning based classification. Comprehensive set of ground truth data were produced by extensive field survey along with very high-resolution UAV images. Preliminary results showed that the pan-sharpened image was better than the multispectral image for detecting the damaged trees. Although the machine learning based classification shows the highest detection accuracy over 80%, it still has some errors. False alarms were found at the ground soil edges and shadows within canopy. The missing error was found on small trees that the crown size was less than the spatial resolution of the multispectral image.