en,
weather
ue SAR
locations
imagery
ing, fire,
analysis
mise the
reter.
From the
ption is
es man-
ita. With
enomena
arbitrary
he event,
equested
mine the
he object
lation of
s helpful
sensing
irance of
3. SAR
sing GIS
ration of
tation of
t
—9
ven mapping
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
2. SAR AND INSAR PRINCIPLE
Airborne or spaceborne SAR sensors provide a 2D mapping of
the scene in the radar frequency domain. Large areas on the
ground are illuminated with radar pulses in an oblique side-
looking manner. Because of the larger wavelength A compared
to aerial images, the SAR signal is usually not attenuated by
scattering with particles in clouds, smoke or even rain. The
choice of a large A (e.g. 70 cm) even offers the opportunity of
penetrating the soil through tree foliage. The range resolution
depends on the signal bandwidth. The distance between the
sensor and a point target is obtained by correlating the received
signal with the complex conjugated transmitted signal [Bamler
and Schaettler, 1993]. With this matched filtering all
contributions of a certain target can be integrated into the
correct range cell. This so-called range compression is carried
out for each range bin of the SAR image.
The angular resolution of a radar antenna of aperture D at a
range r is approximately r-4/D. Hence, for remote sensing
purposes very large antennas would be required for an
appropriate azimuth resolution. This problem is overcome with
the synthetic aperture technique. Along the flight path, many
overlapping radar measurements are carried out and stored.
Each of them can be thought of being collected by one element
of a large synthetic array antenna at the same position. In many
subsequent measurements the signal of a certain point target is
mapped to different ranges cells (range migration). High
azimuth resolution is achieved by a coherent integration of
these distributed signal contributions to the correct image
position. For this purpose, the Doppler frequency is exploited,
which is caused from the relative motion between sensor and
scene.
The characteristics of the received SAR data depend e.g. on
signal properties (wavelength, polarization, transmit power),
object properties (e.g. roughness of surface, dielectric constant
of material), the distance between sensor and scene (the SNR is
indirect proportional to the third power of the range), and the
viewing geometry (e.g. the incidence angle). In case of natural
scenes, the obtained signal is modelled to be the superposition
of the contributions of many independent scatterers inside a
resolution cell. In urban scenes, this assumption does not hold
everywhere, e.g. due to specular reflection and multi-bounce
propagation at building faces and edges. Because of the two-
dimensional sinc-curve of the SAR system impulse response,
dominant point targets appear as bright stars in the image that
may cover the signal of neighboured objects over wide areas. In
general, the mapping of a certain urban area is heavily
dependent on aspect and elevation angles [Dong et al., 1997].
In urban. scenes, the analysis of the SAR scattering matrix
[Guillaso et al., 2003] is useful e.g. to discriminate of man-
made objects from trees or bushes. However, for this purpose
several SAR images of different polarization (e.g. HH, VV and
HV) have to be acquired.
For InSAR processing at least two SAR images taken by
antennas separated by a baseline orthogonal to the flight
directions are required [Rosen et al, 2000]. The different
distances of a ground point to the antennas result in a phase
difference of the received signals from which a DEM can be
calculated. Assuming a constant noise floor, the DEM accuracy
varies locally depending mainly on the SNR, which is estimated
from the coherence (correlation) of the complex SAR images.
SAR data are acquired in the so-called slant range geometry and
have to be orthorectified in a post-processing step. In case of a
magnitude SAR image, an external DEM is incorporated for
precision geocorrection. InSAR data are orthorectified using the
InSAR DEM, whose local varying accuracy has to be
considered for this purpose.
Fig. 2a,b illustrate INSAR magnitude and DEM data covering a
part of the test area (Karlsruhe, city centre and University
campus). Dark areas of the magnitude image (= poor SNR)
coincide with noise in the DEM. The data was acquired from
about 5 km distance in the X-band (A = 3 cm), range direction
is top-down, with off-nadir angle 6 = 57°. The geometric
resolution is about Im in range and 0.75m in azimuth direction.
The reference LIDAR DEM superimposed with map data is
shown in Fig. 2c for comparison.
Fig. 2: a,b) InSAR data in slant range (a) magnitude, b) DEM);
c) LIDAR DEM superimposed with building (red)
and road layer (green).
3. BUILDINGS IN SAR IMAGERY
For discussion of the appearance of buildings in SAR imagery,
the highlighted part of the slant range magnitude image shown
in Fig. 2a was transformed into a ground projection (Fig. 3a).
Usually a flat scene is assumed for geocoding of SAR images,
if no InSAR DEM or external DEM/DTM is available. Fig. 3b
shows for comparison the same area of the LIDAR DEM. Both
images are overlaid with the map data. The extension of the
investigated area is about 400m x 270m.
Buildings appear shifted towards the sensor (range direction is
top-down) in the SAR image, e.g. buildings D and E in the
middle. This is caused from the layover effect: because of the
side-looking illumination and the wide antenna beam width in
elevation, objects located at different positions but with the
same distance to the senor are mapped to the same resolution
cell. At building locations a signal mixture from the roof, the
walls and the ground is the consequence. If no dominant
