erpreted
analyzed
lis: were
y arca to
ts in the
sting no
he study
velength
. poorest
C-band
incident
p image
the HV
ing Was
the best,
incident
the HH
Ma. For
e L-HV
roduced
incident
1. When
ion but
1e large
AR data
incident
ita, and
1s of X-
ie. color
pared to
ral SAR
than the
vo large
> images
number
| ranged
/V LIA
cent for
JV LIA
d the L-
1%) for
| image
mber of
commission errors (7) was found in that image, the X-band VV
polarized LIA image, and the C-band VV polarized LIA image.
It is apparent that VV polarized imagery is of minimum usc for
settlement detection, at least for this environment.
L-band imagery, alone or as a component of a color composite,
consistently produced the highest accuracy. A basic principle
of SAR imagery is that the angle over which an anomolously
large backscatter is observed will increase as wavelength
increases (Raney, in press). In other words, the cardinal effect
is enhanced for longer wavelength systems. This would mcan
that for buildings in a row, the number that will be equal to or
less than 10 degrees from perpendicular to the incident wave
(and the extent of high signal return) will increase as
wavelength increases. For very small villages this increased
backscatter at longer wavelengths would produce a larger,
distinet, bright return on the image than that generated by
shorter wavelength systems. That alone may not be enough to
explain the superiority of the L-band wavelength in settlement
visibility. However, on the L-band (longer wavelength) images
more of the surrounding grass and meadow surface terrain will
appear smooth (dark) than on the shorter wavelength C- or X-
band images due to the surface roughness criterion. For the
latter images the differences in response and contrast between
settlement structures and grasses would be less, producing a
more homogeneous appearing landscape. The L-band image
would generate a high backscatter response {rom a larger arca
of the settlement against a surrounding arca of low return. The
only other arcas with responses similar to settlements would be
some forested and wetland areas, and they would bc
distinguishable from settlements by different shape and
texture. The results of this study support those observations.
The fact that the best single image was HV L-band supports
earlier work (Bryan, 1975; Haack, 1984; Henderson and
Mogilski, 1987; and Lewis, 1968) that cross-polarized imagery
provides more separation of human settlement structures versus
vegetation due to the role of volume scatter. The accuracy of
the total power images, unavailable to the earlier researchers,
suggests that they may produce equal or better results than
cross-polarized imagery. Further investigation of total power
images merits attention.
The superiority of large incident angle images also supports
carlier work in similar terrain (Henderson, 1995; Kessler,
1986). Whether better accuracy would be attained with a
slightly smaller incident angle than that used here (57.7
degrees) remains to be addressed.
Each of the color composite images produced higher accuracy
than any of the separate component images. Although some
settlements were visible on only one or two of the individual
mages, many settlements on the color composite were not
detected on any of the single band component images. Similar
to many other remote sensing applications, the "multi-" aspect
of multispectral SAR imagery is very important in settlement
detection and delimitation of the built-up area. Each
wavelength may be sensitive to select backscatter components
in urban/settlement arcas. Identification of these factors is the
289
subject of current rescarch. The color composites also enhance
settlement visibility by providing improved contrast and color
differences the interpreter uses to separate human settlement
(bright white returns) from natural elements of vegetation,
soils, and water (colors and black).
SUMMARY
A small area of a Europcan northern boreal environment served
as the basis of this study. For such an environment it has been
shown that L-band imagery is superior to X- or C-band
imagery for settlement detection. Cross-polarized imagery is
the most accurate but total power images may be equal or
better. More comparison of and work with total power images
arc requisite. Large incident angle images are quite superior
to small incident angle images. Use of multispectral SAR
composites is the most accurate of all images, but the exact
combination of component images that will produce thc
maximum accuracy remains to be determined. Future work
will report on population size, population estimate, and
scttlement infrastructure variables.
BIBLIOGRAPHY
M.L. Bryan: "Interpretation of an Urban Scene Using Multi-
Channel Radar Imagery”, Remote Sensing of Environment,
4(1), 1975, pp.49-66.
B. Brisco, F.T. Ulaby, M.C. Dobson: "Spaceborne SAR Data
for Landcover Classification and Change Detection",
Proceedings of the IEEE International Geoscience and Remote
Sensing Symposium, San Francisco, 1983, pp.1.1-1.8.
B.C. Forster: "An Examination of Some Problems and
Solutions in Monitoring Urban Areas from Satellite Platforms”,
Int. J. of Remote Sensing, 6(1), 1985, pp.139-151.
B.N. Haack: "L- and X-Band Like- and Cross-Polarized
Synthetic Aperture Radar for Investigating Urban
Environments", Photo. Engineering and Remote Sensing,
50(3), 1984, pp.331-340.
F.M. Henderson: "An Analysis of Settlement Characterization
in Central Europe Using SIR-B Radar Imagery", Remote
Sensing of Environment, 1995, in press.
and M.A. Anuta: "Effects of Radar Systems
Parameters, Population, and Environmental Modulation on
Settlement Visibility", Int J. of Remote Sensing, 1(2), 1980,
pp.137-151.
and K.A. Mogilski: "Urban Land Use Separability as
a Function of Radar Polarization", Int. J. of Remote Sensing,
8(3), 1987, pp.441-448.
and Z. Xia: "Radar Application in Urban Analysis,
Scttlement Detection and Population Estimation", Chapter 15
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
