ttn > _ mma 
Figure 6 presents the results of the application of the Fourier bandpass 
classifier to the central part of the SIR-A image of Belize. This area was 
resampled so that pixels represent 100X100 m areas. The classification map 
shows that the mountainous area Is easily distinguished from the smoother : B 
lowland areas. The textural signature of this unit (unit 2) shows large 1982, 
variance at all frequencies with a slight drop in variance at higher scatte 
frequencies. The smooth granitic areas are also well delineated as unit 5, 
and have low variance In all the bandpasses, with a slight rise at higher B 
frequencies. The map does not discriminate well between the karst and Its discri 
surroundings. Most of the karst area is classified in unit 1 which has an 
intermediate variance that steadily rises with increasing frequency. The D 
karst is often misclassified, however, as unit 3 which has a similar Seasat 
signature. A better selection of band widths and positions may provide better 
discrimination between the karst and other smooth areas. H 
Proc. 
The Fourier bandpass classifier was also applied to the same area in the 
Seasat image of Belize (Figure 7). This area was also resampled to a 100X100 e 
m pixel size. The results are nearly identical to those for the SIR-A Image. sensin 
The mountainous area is well-delineated, while the smoother karst, granitic, 
and marine sediments are classified into the same unit. The mountainous areas P 
have the same type of textural signature as in the SIR-A map: relatively high Assoc. 
variance, with a slight drop-off at higher frequencies. Similarly, the karst 
and other smoother areas of unit 1 have a similar signature as in the SIR-A 
map: rising variance with increasing frequency. Differences In the Seasat 
map include the discrimination of the mountainous area into two units (2 and 
3). This corresponds to a distinction between the metamorphic rocks, mapped 
as unit 2, and the continental sediments, mapped as unit 3. Another 
difference In the Seasat-derived texture map Is the Inclusion of the higher 
relief karst areas into unit 2. This is a result of the higher sensitivity of 
the Seasat system to low slope angles. Some of the misclassification on this 
map may also be improved with a better choice of bandpasses, especially at 
higher frequencies. 
CONCLUSIONS 
The techniques discussed here are being developed to aid in the geologic 
interpretation of SAR images. The application presented here is In an area of 
heavy tropical vegetation, where very little other data can be obtained 
directly from remote sensing images. As understanding of the relationships 
between Image texture, topography, Iithology, geomorphology, and climate 
Improves, textural information from SAR images may be used for the 
discrimination of units as well as the identification of rock types. An 
active program is being pursued at JPL to integrate textural Information from 
radar images directly with backscatter data from those same Images, and with 
compositional information derived from visible-near Infrared sensors such as 
Landsat (eg. D.L. Evans, this volume). The role of quantitative textural 
information in this type of multisensor analysis promises to be significant. 
ACKNOWLEDGEMENTS 
This paper benefitted from discussions with M. I. Daily, Ph. Rebil lard, 
W.D. Stromberg, and especially D.L. Evans. Image processing was done at the 
Image Processing Laboratory, JPL. The research described here was carried out 
at the Jet Propulsion Laboratory, California Institute of Technology, under 
NASA contract NAS 7-100. 
264 
pr till «il Bl