Figure 5. Simulated image overlaid with PS 
Figure 6. Oblique view aerial image 
In: Wagner W., Székely, B. (eds.): ISPRS TC VII Symposium - 100 Years ISPRS, Vienna, Austria, July 5-7, 2010, IAPRS, Vol. XXXVIII, Part 7B 
Figure 4. Simulated image containing single and double bounce 
contributions 
3.3 SAR simulation for PSI analysis 
In order to compare the PS extracted from the data stack with 
the results of the simulation, the simulated image is warped to 
the geometry of the data stack. This is necessary due to different 
coordinate systems of the simulated scene and the real scene, 
and because of small geometrical errors caused by the 
simulation of an airborne sensor system. 
We applied an affine transformation for the warping procedure. 
The coefficients of this model were estimated using a number of 
six tie points distributed over the whole simulated scene. The 
result is displayed in Figure 3 and Figure 4. The mean 
amplitude image of the stack is displayed in Figure 3, while the 
simulated scene containing singe and double bounce 
contributions is shown in Figure 4. It can be clearly seen, that 
the main structures of the buildings are reproduced by the 
simulation. 
Within the simulation we can distinguish between single- and 
double-bounce reflections. Locations of double-bounce between 
the dihedral comer reflector spanned by building walls and the 
ground in front are clearly visible in the real and the simulated 
SAR image. 
However, the details of the façades are not visible, even facade 
structures which cause very strong reflections are often too 
small to be represented in the 3D model. 
The PS set superimposed on the simulated image is shown in 
Figure 5. It is clearly visible, that besides some geometrical 
inconsistencies (see skyscraper at the top of the image) the PS 
set matches fairly well with the simulated image. On the other 
hand the main problem of the whole approach shows up. Since 
most of the PS are generated by small scale building features 
(like the above mentioned balconies and windows), which are 
not modelled in the simulation, detailed analysis of the physical 
nature of the PS is virtually impossible. 
The whole situation can be best illustrated by considering the 
part of Figure 5 marked by the red rectangle. In this area lots of 
PS appear, but the simulation just indicates a homogeneous 
area. These PS reside mainly on the roof of the round shaped 
building in the centre of the building block, as can be seen from 
the height data in 
Figure 2. An oblique view aerial image shown in Figure 6 
reveals the structures leading to this group of PS. First of all, a 
lot of hardware, which may be used for ventilation purposes, is 
visible. Additionally, a metallic frame surrounds the dome-like 
part of the roof in Figure 6. Both types of structures are likely to 
produce PS, but are not contained in the 3D model used for 
simulation. Therefore an assignment of the respective PS to 
these building features using the shown simulation results is 
hardly possible. 
For simulating these structures we would need a 3D building 
model reconstructed from terrestrial laser scanning or close- 
range photogrammetry. Less accurate, but for some applications 
still acceptable, would be models reconstructed using façade 
grammars describing the façade (Becker 2009). In this way, 
models of simple standardized buildings can be generated 
without using high-resolution laser scanning data or close-range 
photogrammetry. 
The Sony-Center and the surrounding buildings used in our 
experiments are not standardized buildings easily representable 
in a façade grammar. We also do not have high resolution 
façade data. Therefore, we couldn’t reconstruct a 3D building 
model in the required quality. 
However, the simulation is useful to retrieve information about 
the PS within the green box. We know that virtually all 
scattering within this area is due to the roof of the building, 
which encompasses the round shaped one.