An area of central Belize (Central America) was chosen as a site to test
several techniques of texture analysis using radar images produced by both
SIR-A and Seasat (Figures 1 and 2). The area is characterized by a nearly
continuous vegetation canopy and a cloudy, tropical climate (no cloud-free
Landsat image of the area could be found). A variety of rock types are
present in the area (Figure 3). Resistant metamorphic and continental
sedimentary rocks make up a mountainous area in the center of the image, while
the very smooth area (area 4, Figure 1) Is underlain by granitic rocks which
weather quickly in the tropical climate. The peculiar "bumpy" texture of area
2 (Figure 1) is representative of karst topography formed by solution of
Cretaceous | imestones. Note the enhancement of the small slopes in the
granitic and karst areas, and the distortion of the mountalnous slopes in the
Seasat image (Figure 2) relative to the SIR-A image (Figure 1).
TECHNIQUES FOR TEXTURE ANALYSIS
A. Split-spectrum
A simple technique for textural analysis of radar images developed by
Daily (1982) splits the spatial frequency spectrum produced by a Fourier
transform Into two parts. The high frequency part was dominated by
topographic variations, and the low frequency part was related to broad-scale
backscatter variations in the area studied by Daily (1982). The low frequency
component was then coded in color, while the high frequency component was used
to modulate the intensity of the color. This produced a color image that
enhanced broad-scale backscatter variations with color while retaining
topographic information in the usual way. This technique was found to be most
useful where there was a clear distinction between smal |-scale topographic
variations and larger-scale backscatter variations.
In a heavily vegetated region such as central Belize, both high and low
spatial frequencies are produced by topographic variations, as explained in
the introduction. Thus, the split-spectrum technique provides little new
information, although It was possible to differentiate the higher frequencies
of the karst and marine sediments from the mountainous, low frequencies of the
metamorphic and continental sedimentary rocks.
B. Fourier transforms of subareas
In order to study details of texture in small areas, Fourier transforms
can be obtained for subareas of an image. Figure 4 shows three subareas
extracted from the SIR-A image of Belize (Figure 1). These areas are 512X512
pixels in size (pixels are 25X25 m in the SIR-A data). Next to each subarea
in Figure 4 is Its 2-dimensional Fourier transform. The Fourier transforms
are presented in a polar coordinate system with the lowest frequencies at the
center, and frequency increasing radial ly outward. The azimuthal coordinate
corresponds to azimuth in the Image. The brightness of a point is related to
the variance or "power" at that frequency and orientation. Each transform was
scaled independently, in order to minimize saturation in the pictures, and
hence should be compared on a qualitative basis.
Geologic interpretation of the information contained in Fourier
transforms of radar images has received little attention. Pike and Rozema
(1975) used similar, one-dimensional techniques to study topographic profiles
over various sites of differing geology, climate, and geomorphic history.
They found that: a) the integrated variance ("power") spectrum is a good
measure of overall roughness, b) the slope of the variance spectrum in log-log
coordinates was proportional to the relative strengths of large- vs.
small-scale topography, and c) topographic periodicities were observed as
262
A a [1 0 0 NN CAIN EOD C A
— ae teens OV
peaks
resist
factor
The s
Throug
draina
Jointi
1
presen
rate o
The 1
examp |
a rap
relati
resist
recent
probab
also c
in its
with t
Transf
smal |
that
hourgl
This
mounta
7
same
caused
Introd
fold-o
images
|
the Se
(Figur
peak c
howeve
fol d-o
to con
C
D
geol og
desira
a: lar
was de
images
filter
bands
are th
algori
signat
varian
Evans,
