Abstract
We proposed a novel index, NWI, to detect oak wilt, which has been wide-spreading around Japan Sea seaboard of the main island of Japan, from remotely sensed visible/near-infrared hyperspectral data. At first, we estimate differences in composition between oak wilt and healthy oak leaves based on pure spectral data collected by leaf level measurements. Subsequently, we characterized the spectrum by two distinctive features based on three narrow bands of green, red, and near infrared, i.e. i) NDGI as a normalized difference between green and red band and the numerical, ii) (NDVI+NDGI) as an approximation of the second-order central difference at red band. The NWI, defined by the product of -NDGI and (NDVI+NDGI), indicates positive value only in case of oak wilt leaves. The application of the NWI to remotely sensed airborne hyperspectral images provides a high accuracy extraction of oak wilt with an Az value from conditional ROC analysis of 0.95.
