Spectral Unmixing of Landsat TM Imagery for the Detection of Dutch Elm Disease in Native Elms in Saskatchewan

摘要

Landsat Thematic Mapper (TM) imagery from August 18th, 1991 was atmospherically and geometrically corrected using standard correction techniques. Supervised and unsupervised classification were not successful in identifying any positive Dutch Elm Disease (DED) sites (i.e., early or late stages of DED). The process of spectral unmixing was attempted using spectra for healthy elms along with early and late stages of DED onset. Either early or late DED was detected at all 123 test sites in Sherwood Forest study area using this method. Spectral unmixing appears to be the best method for the identification of DED from mixed Landsat TM pixels provided the location of the elms are already known.