Environmental Engineering Reference
In-Depth Information
FIGURE 5.3
A cluster of skyscrapers in the Shanghai, China, Business District as seen in a TerraSAR-X image.
Depending on the viewing direction, acquisition of a large
urban scene can be plagued with corner-cube or multi-bounce
phenomena (Dong, Forster and Ticehurst, 1997); regions of radar
shadow (e.g., cast behind buildings) coincide with noisy InSAR
elevation data. One may overcome such obstacles by acquiring
SAR data from different vantage points and fusing interpretation
results (Michaelsen, Soergel and Thoennessen 2005; Soergel
et al
., 2005) or fusing InSAR with other 3D data such as LIDAR
(Gamba, Dell'Acqua and Houshmand, 2003; Stilla, Soergel and
Thoennessen, 2003). In any case, the appearance of buildings in
radar images is complex enough to make perceptual approaches
a good option (Michaelsen, Soergel and Thoennessen, 2006), as
well as gestalt perception rules (Michaelsen, Middelmann and
Sorgel, 2007). Figure 5.3 shows how a cluster of skyscrapers
from the centre of Shanghai is represented in a TerraSAR-X
spotlight image, which gives an idea of how necessary it is
to consider approaches capable of some degree of modelling
and interpretation. Just to highlight a few of the problems one
encounters, the reader is invited to not the following:
In optical images a closer object occludes and thus cancels
completely another object behind it; in radar images the location
of backscatterers is basically determined by the optical path
length, and thus structures in different locations may appear
overlapping because their line-of-sight distances from the sensor
are similar. So do the two skyscrapers at the top of Fig. 5.3.
For the same reason, double or triple bounces make the
optical path longer that would be for line-of-sight acquisition,
and thus objects illuminated by mirror-reflected rays from a
building appear mislocated. Such phenomenon is not clearly
visible in Fig. 5.3.
In optical images, bright spots may cause saturation of the
sensor and thus results in an underestimation of local reflectivity;
in radar images, due to the impulse response of the SAR sys-
tem, strong backscatterers produce an overestimation of radar
response over a number of pixels along the same row and column
where the backscatterer is located. This phenomenon shows as
''bright crosses'' and it is visible in Fig. 5.3 at the right end of one
skyscraper in the middle of the image.
When the complexity of the problem itself is the main issue,
the adoption of techniques developed for airborne radar on
spaceborne data can be a natural choice. Also, papers specifically
devoted to spaceborne data (Franceschetti
et al
., 2008) can now
be found, presenting new applications unsuitable for airborne
radar (Suchandt
et al
., 2008). Extraction of electromagnetics
instead of geometric properties of the buildings seems to be at
the research forefront (Guida
et al
., 2008).
For building reconstruction, the greatest benefits of space-
borne VHR seem to derive from data availability (all-weather
operation, inherent repeatability, large-scale acquisition) rather
than from data characteristics. Although the ground resolution
(1 m) would allow discriminating similarly-sized details in the
scene, the actual situation shows images that do not allow easy
extraction of the buildings, as shown in Fig. 5.4. Here, the his-
torical centre of L'Aquila, Italy, is represented, and it is clear how
difficult it is, not only to exactly locate the backscatter contribu-
tions presented in Fig. 5.1, but even in some cases to separate
neighboring buildings.
5.3.2
Damage assessment with
VHR SAR data
The use of radar data for damage assessment was proposed
long before high-resolution satellite SAR data became available.
Imhoff
et al
. (1987) used SRTM data to determine flooded areas
in Bangladesh and to inventory affected assets based on a land
cover classification derived from Landsat data. Though, more
difficult goals as earthquake damage assessment took longer to
Search WWH ::
Custom Search