Integrating spectral, spatial, and terrain variables for forest ecosystem classification

Sets of spectral, spectral-spatial, textural, and geomorphometric variables derived from high spatial resolution Compact Airborne Spectrographic Image r (CASI) and elevation data are tested to determine their ability to discri minate landscape-scale forest ecosystem classes for a study area in norther n Ontario, Canada. First, linear discriminant analysis for various spectral and spectral-spatial variables indicated that a spatial resolution of appr oximately 6 m was optimal for discriminating six landscape-scale forest eco system classes. Second, texture features, using second-order spatial statis tics. Significantly improved discrimination of the classes over the origina l reflectance data. Finally, addition of terrain descriptors improved discr imination of the six forest ecosystem classes, it has been demonstrated tha t, in a low- to moderate-relief boreal environment, addition of textural an d terrain variables to high-resolution CASI reflectance data provides impro ved discrimination of forest ecosystem classes.