ul 2004 
y their 
there is 
ngs. For 
ich also 
ver. 
ON 
Juisition 
ver and 
For this 
st effort 
a shown 
Meier et 
sampled 
detected 
ction of 
uildings 
ns were 
a and 
ve been 
les a 
t 
OW 
building 
without 
Tease to 
ation of 
  
without 
sr cent. 
Fig. 3b 
ried out 
he other 
altitude 
quisition 
ring e.g. 
and £. 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
       
3 7) : M: 
ii Pip 
Fig. 5: a) simulation result for given SAR parameters 
(illumination from top): layover (white), shadow 
(black) layover and shadow (dark gray), and reliable 
data (bright gray) b) simulation for illumination 
from left. 
^ 
5. ANALYSIS OF SAR DATA OF DENSE URBAN 
SCENES 
The interpretation of SAR imagery in dense urban scenes 
without context information is often difficult even for experts. 
An important task in case of disaster monitoring is the detection 
of changes, e.g. damaged buildings and infrastructure. 
However, context information is usually not available for all 
areas at the same level. The benefit from integrating GIS data in 
the analysis will be discussed for different levels of available 
data in the GIS: 
5.] 2D map data 
In the simplest case the availability of a vector map containing 
a layer with building footprints is assumed. According to the 
sensor parameters, the building footprints can be transformed 
into the SAR images. The lack of height information leads to a 
shift of the transformed footprints similar to Fig. 3a. However, 
due to this shift the layover regions at the buildings are clearly 
visible. These areas could be e.g. excluded from the calculation 
of the building height from the InSAR DEM (Fig. 6b). The 
road layer of a vector map is useful for the discrimination of 
roads from cast shadow of buildings. 
503 
5.2 2D map data and mean building height 
Using height information, the accuracy of the projected 
footprint positions is enhanced. Here, the mean height was 
determined by averaging the LIDAR data inside the building 
footprints. Fig. 6a shows the transformed footprints 
superimposed on the slant range magnitude image. The 
footprints of buildings, e.g. building D and E, are now correct. 
Due to the height difference, the two parts of building 1 appear 
at the same location in the SAR image. 
The change detection capabilities for urban structures from 
SAR images can be demonstrated e.g. with buildings B and C. 
Some parts of these buildings are missing in the map (Fig. 3b), 
because they have been built after the map production. At the 
related SAR image positions, some building structures can be 
identified. 
With the method described in section 4, layover and shadow 
areas can be predicted for the slant range geometry as well. The 
predicted shadow is useful for the discrimination between the 
cast shadow from buildings and other low backscatter areas like 
roads. By intersecting the road layer and the predicted layover 
areas, those bright objects on the road not coinciding with 
layover are hints to vehicles. 
A further step is a simulation of SAR data. Because SAR 
images depend on the scene geometry and material properties, a 
complete simulation should address these features 
[Franceschetti et al., 1992]. But, usually detailed information 
about the materials e.g. of building roofs is not available. 
However, the image interpretation can benefit from simulations 
based on geometric properties alone. Fig. 6c shows the 
simulation result for the magnitude image. The appearance of 
the simulated and the acquired images differ noticeable, 
because besides the material properties the roof structures and 
the vegetation are not considered. Nevertheless, layover and 
occlusion areas are clearly visible. 
5.3 LIDAR DEM 
In order to reduce the deviations between the simulated and the 
acquired data, additional context knowledge is required. A high 
resolution LIDAR DEM provides information about the roof 
structure of the buildings. Furthermore, the vegetation impact of 
the object visibility in SAR data can be considered [Soergel, 
2003a]. The wavelength dependency of the volume scattering 
inside tree or bush foliage can be taken into account by the 
choice of the suitable LIDAR data mode. For radar signals with 
short wavelength, first-pulse data is appropriate, while last- 
pulse data is advantageous in case of a larger radar wavelength. 
The simulation result depicted in Fig. 6d was based on the first- 
pulse DEM shown in Fig. 3b. 
Comparing Fig. 6c and 6d, the influence of vegetation on the 
SAR mapping seems to be large even in urban areas. The signal 
contribution caused by the geometry of roof structures is now 
considered. The impact of material properties on the backscatter 
cannot be taken into account. Therefore, some buildings and the 
place in the middle appear darker in the acquired data. For 
example, a flat roof made of concrete shows up black in the 
magnitude image, because the signal is reflected away from the 
sensor at the smooth surface. Furthermore, the influence of 
dominant scattering is not covered with this kind of simulation. 
However, an image interpreter may benefit from such a 
visualization, e.g. for spotting damaged buildings.