contributions have boon and aro made by rosoarch institutions in Canada and the USA, e.g. 
the University of Kansas, E1UM, JPL and CCltS. There are also some contributions by euro- 
pean institutions, mostly initiated by ESA. The following summary tries to sample relevant 
publications. 
A number of authors (e.g. [Li 80, Goodenough 80, Guindon 80, Ulaby 82]) argue that the com 
bination of radar images with optical/thermal imagery will yield remarkable improvements 
of classification results as compared to the evaluation of data of a single sensor. [Brisco 83] 
compared in detail the applicability of SEASAT, SJR-A and LANDSAT imagery alone or in 
combination to separate land use classes. Other investigators favour the evaluation of multi 
ple radar imagery alone. [Ulaby 80] used dual polarized imagery of the ERIM L-band radar 
sensor to classify successfully different vegetation types. [Brisco 82] evaluated multitemporal 
imagery of the ERIM 4-channel radar sensor to generate land cover classifications. Finally, 
[Frost 85] described in detail all processing steps for a land cover classification using the 
maximum likelihood algorithm. 
All authors refer to radar speckle as unwanted noise which had to be reduced or eliminated 
by application of multilook techniques, or well known lowpass or specially developed filters. 
In Europe there were two radar remote sensing campaigns, namely the SAR-580 campaign 
in 1981 which produced multiband, multipolarized radar imagery, and the AGRISAR cam 
paign in 1986 which produced X-Band, multipolarized, multitemporal imagery. [Curran 84] 
argued that SAR imagery may not be very useful for land cover classification tasks because 
different, land use classes show identical backscatter coefficients and, on the other hand, soil 
moisture has a bigger influence on the backscattered intensity than different types of land 
cover. [Nueesch 84], however, proved the separability of 5 different crops in SAR-580 radar 
imagery of a. tcstficld of 2 kin x 2 km on the basis of statistical texture measures. [Churchill 84] 
reported that 5 different tree species, urban areas, grassland, sommer cereals, and root crops 
were separable in SAR-580 imagery on the basis of a statistical classification of the image 
intensities. The following statements may be concluded from these publications: 
• several approaches to the evaluation of SAll imagery have been developed to solve many 
different tasks and requirements; no consensus in the different methods used nor in the 
assessment of the potential of radar remote sensing has been reached; 
• basic problems of the radar backscatter process are still to be solved; 
• most of the investigations used small or very small test areas; large area land cover 
classification results have not been reported yet; 
• most of the investigations used features and methods developed for the evaluation of 
imagery from optical or thermal sensors; special features to describe characteristics of 
the radar signal seem to be missing yet. 
Proposal of Special Texture Features 
A number of different, methods to define and measure texture in imagery has been developed 
([Wechsler 80]), and they have been applied successfully to analyse imagery from different 
sensors. Most widely used are statistical texture parameters ([Ilaralick 78]), which are based 
on calculations of intensity differences. The radar signal, however, is not only disturbed by 
system noise, but also by the speckle phenomenon; hence, the intensity value at each pixel 
in a SAR image is uncertain to a higher degree than in any image of the optical spectrum. 
Pixel intensities and intensity differences will, therefore, present no reliable basis for feature