Proceedings Article | 30 December 1997
Proc. SPIE. 3222, Earth Surface Remote Sensing
KEYWORDS: Landsat, Data modeling, Satellites, Reflectivity, Vegetation, Satellite imaging, Light sources and illumination, Radiometry, Earth observing sensors, Spectral models
Linear spectral mixture models (SMM) with image endmembers (IEM) and with reference endmembers (REM) were tested for discriminating maritime pine stands and shrublands in a Landsat-TM image of Central Portugal. For both types of EM, IEM and REM, two types of SMM were tried: SMM with three EM (SMM-3), i.e., green vegetation, soil and shade, and SMM with five EM (SMM-5), where the EM were the components of the landscapes that we were interested on, i.e., pine canopy, shrub, soil, forest litter and shade. Results showed that in the SMM-5, REM need to be used, since IEM were not pure enough. We verified that in the SMM-5, there was not a single set of EM that could be applied to the whole study area, because the shrubs that exist underneath the pine canopy and in the shrublands could not be modeled just by using a shrub EM. Therefore, SMM-5 require a multi-endmember approach, where the set of EM may change from pixel to pixel. In the SMM-3, an accurate discrimination of shrublands and pine stands (90% accuracy) was achieved by thresholding the shade fraction. In these simpler SMM, IEM and REM produced similar results.
