Fig. 6: Fractions of the endmember STONE PAVEMENT (top) and NUMMULITE-COVERED SURFACES (bottom) (caption 
See Fig. 5) 
By the applied method 98% of the pixels in the study area 
could be unmixed with a rms «29^ (Fig. 7). 
The average rms of the whole image was 0.796, i.e. slightly 
larger then the noise level of 0.5696. Pixels with a large 
rms occur mainly in the surrounding of the dunes and 
along the slopes. In these areas the intensity of the incom- 
ing sunlight varies due to the relief. 
230 
5 CONCLUSIONS 
The spectral mixture analysis provides a fractional classi- 
fication of the LANDSAT-TM image for bare soil surfaces 
in the study area. The application of the field spectroscopy 
was important for the determination of the endmembers 
and makes it possible to keep their number as small as 
possible. The spatial resolution of the LANDSAT-TM de 
mands as an endmember shadow. the fraction of shadow 
provides information about the relief and the surface 
roughness. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996