to test 
| by both 
| nearly 
oud-free 
pes are 
tinental 
e, while 
ks which 
| of area 
ution of 
in the 
S in the 
oped by 
Fourier 
ted by 
ad-scale 
requency 
was used 
ge that 
etaining 
be most 
ographic 
and low 
alned in 
tle new 
juencies 
5 of the 
insforms 
subareas 
512X512 
subarea 
insforms 
; at the 
rdinate 
ated to 
Orm was 
es, and 
Fourier 
Rozema 
rofiles 
istory. 
a good 
log-log 
- VS. 
rved as 
a 
peaks in the variance spectrum. The integrated variance can be related to the 
resistance of the rocks underlying the area, its tectonic history, or climatic 
factors. Thus, it may be related to the geomorphic "maturity" of an area. 
The slope of the spectrum is also probably related to geomorphic maturity 
Through the effects of the development and integration of small-scale 
drainage. Periodicities in the variance spectrum can be the result of 
jointing, bedding, or other, more exotic factors. 
These kinds of interpretations can be applied to the Fourier transforms 
presented in Figure 4. One difference in the transforms is apparent in the 
rate of drop-off in brightness away from the center. This is proportional to 
the texture scale in the same way as the slope of the variance spectrum. For 
example, the large-scale texture of the mountainous subarea (subarea 5) causes 
a rapid drop-off, while the smoother granitic area (subarea 4) has a 
relatively "flatter" transform. These observations imply that relatively 
resistant lithologies make up the mountainous area, or that it was uplifted 
recently. In contrast, the relatively low relief of the granitic area is 
probably a result of rapid weathering in this climate. Directional features 
also cause differences in the Fourier transforms. Subarea 4 includes a ridge 
in its corner, and the transform of that area shows a strong directional trend 
with that orientation. Periodic topography is also observed in one of the 
transforms. The karst of subarea 2 is periodic, and this is expressed as 
small bright spots to either side of the center in the Fourier transform of 
that area (Figure 4b). The Fourier transforms of these areas all have an 
hourglass shape caused by the enhancement of slopes facing the radar sensor. 
This is more pronounced for the steeper and more numerous slopes of 
mountainous areas. 
The "hour-glass" effect is also pronounced in Fourier transforms of the 
same subareas extracted from a Seasat image of Belize (Figure 5). This is 
caused by the enhancement of slopes by the Seasat system, as explained in the 
Introduction. The enhancement of small slopes in subareas 2 and 4 and 
fold-over in subarea 5 is readily apparent upon comparison of the Seasat 
images of the subareas (Figure 5) with those from SIR-A (Figure 4). 
In general, similar information is provided in the Fourier transforms of 
the Seasat subareas. More structure is apparent in the transform of subarea 4 
(Figure 5d), while the transform of the karst area (Figure 5b) barely shows a 
peak caused by periodicity. The transform of the mountainous area (Figure 5f) 
however, drops off rapidly, probably because of the dominant effects of 
fold-over. In this case, the transform of the SIR-A image (Figure 4f) appears 
to contain more useable information. 
C. Spatial frequency bandpass classification. 
Detailed study of small subareas can lead to better understanding of the 
geologic and geomorphic factors operating in small areas, but it is also 
desirable to produce a map of geologically meaningful textural variations over 
a large area. One technique that can be used to present textural variations 
was described by Blom and Daily (1982). The technique involves production of 
images that represent only certain scales of texture. This is accomplished by 
filtering the Fourier transform of an area into several spatial frequency 
bands and producing an image for each of these bands. These bandpass images 
are then used in a standard, unsupervised "multispectral" classification 
algorithm to produce a map, the units of which represent unique textural 
signatures. One could also use [images in which pixels represent local 
variance, or other statistical quantities, in different size boxes (eg. D.L. 
Evans, this volume) in the classification algorithm. 
263 
tn ND P SI SSR 
"s