Very-high-resolution spaceborne synthetic aperture radar and urban areas: looking into details of a complex environment - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 63

FIGURE 5.4 A portion of the historical centre of L'Aquila as seen by the COSMO/SkyMed SAR in spotlight mode. The original
radar data was generated by the Italian Space Agency, and provided by the Italian Civil Protection Department.

be addressed. Generally speaking, damage assessment is a change
detection problem; in radar images, the changes in the observed
objects manifest themselves in intensity changes (a feature shared
with optical images) and decorrelation of scatterers (peculiar to
radar images).
Decorrelation due to the change in orientation of strong
scatterers was considered by Usai and Klees (1999), and Mat-
suoka and Yamazaki (2000). Change in amplitude characteristics
generated a different investigation stream (Aoki, Matsuoka and
Yamazaki, 1998; Matsuoka and Yamazaki, 2004). A comparison
between the two approaches (orientation of strong scatterers vs.
change in amplitude) was attempted by Yonezawa and Takeuchi
(1999). Apparently similar behaviours were reported, although
the actual correlation between the damage level and the data
features was not outstanding.
Multitemporalstacksofimageswereproposedasawayto
improve accuracy of results (Trianni and Gamba, 2008), as well
as the use of ancillary information (Gamba, Dell'Acqua and
Trianni, 2007). Further improvement of the results is expected
from the much larger amount of information conveyed by VHR
SAR images, also because of the significant increase in the number
of samples in the statistics.
For example, one may use statistical knowledge to develop a
different approach for damage assessment, that is investigation of
post-event statistics only, rather than classical change detection.
This approach makes sense given the young age of spaceborne
VHR SAR systems, for which the absence of any pre-event image
at a given location is not an unlikely event.
Building on the idea developed by Gamba, Dell'Acqua and
Trianni (2007) of exploiting ancillary data, this approach was

tested on a real case, namely, the 6 April 2009 Italy earthquake in
L'Aquila, Italy.
Visual inspection of the Google Earth image allowed produc-
ing a GIS layer with 58 city blocks, as shown in Fig. 5.5; the average
block size is 0.1176 km 2 , where the smallest polygon is 0.0146 km 2
and the biggest one is 0.698 km 2 . Every block in the city centre
contains 100 to 150 buildings. A comparison of each block with
a layer containing footprints of severely damaged buildings was
performed. These latter buildings were visually extracted from
post-event aerial images acquired by the Italian Air Force and
kindly provided by the Italian Civil Protection Department. Each
block was marked with the percentage of its area covered by dam-
aged buildings (DAR damaged area ratio), a number between

0 and 46.4%, with an average value of roughly 4%. Of the total
58 blocks, 37 reported DAR = 0. A number of different texture
measures were computed on a post-event, amplitude spotlight
COSMO/SkyMed image over the same area, geocoded terrain
corrected, provided by the Italian Space Agency through the
Italian Civil Protection Department. Every single texture map
was averaged over each polygon in the series, creating a series of
58 average texture values per each selected texture.
Pearson correlation coefficients were then computed between
the series of DAR values and each of the texture average values.
Most correlations, namely between DAR and data range, mean
and entropy, were found to be negligible (less than 0.1); yet,
correlation with variance was significantly higher, namely 0.275
with a window size of 3 × 3 pixels, decreasing to 0.246 at
a window size of 19 × 19 pixels. Similar levels of correlation
were found in analyzing COSMO/SkyMed data on Guan Xian,
China (Dell'Acqua, Lisini and Gamba, 2009), after the Sichuan

=