anbul 2004
rent seal-
ion of the
' instance,
n we may
he sealing
proposed
resolution
pling, and
th respect
iore bands
001), for
level co-
n and sum
ed to im-
ound that
land tex-
ventional
\bination.
yrmed di-
nation. À
7), where
tion. The
roves the
y degrad-
)), where
sification
ses. The
x 3 win-
1 filtered
ome sort
spectral
us para-
x density
we may
exempli-
of textu-
ronment
vork the
ilt.up to
medium
ificantly
ause we
stead of
windows
1 respect
ace Soil
different
tance in
> classes
1a man-
bers for
method
rial and
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part Bl. Istanbul 2004
residential zones to be clearly defined, mainly based on the
relative amount of vegetation cover.
Presently the sealed area in urban areas can be mapped to
a sufficient degree of accuracy by using textural features
in conjunction with spectral information. It is still to be
investigated what the critical issues are when this approach
is applied on a large scale project, on more than one or
few test sites. Moreover, automatic choice of the optimal
window size for the measures as well as of the measure
set need to be related to the scale of the *objects" we are
looking for in the urban environment. These may be the
lines for future research on this matter. In the meantime,
current methods combining spectral and textural features
may be sufficient, or even only spectral features like those
that allow to calculate the NDVI, on which the method by
the GUS service provider is based, followed by correction
by a remote sensing expert.
In conclusion, main limitations of the current algorithms
for sealing mapping are that the estimation of the sealing
degree based on very few sensed quantities, where textural
information is not considered, while, where textural infor-
mation is considered, there is actually a lack of extensive
testing. Research lines that should be addressed to solve
these problems are therefore:
e automatic definition of the window size where textu-
ral features are used;
e extensive testing of techniques combining spectral and
textural features over a number of different sites with
different characteristics.
4 RESULTS AND MAPS
As an example of the above mentioned products, we offer
in this section the results of a recently developed method-
ology for the extraction of urban area information from
medium resolution SAR satellite data. We focus in par-
ticular on RADARSAT data, as a suitable mean to under-
stand to which extent this mapping approach may be use-
ful with the finest spatial data now available. This, in turn,
may be a first guess of what we may expect from finer res-
olution, Low Earth Orbit satellites, like Cosmo/Skymed,
TerraSAR.x And SARLupe, as well as RADARSAT-2.
It has been shown in Dell' Acqua and Gamba (2003) that
interesting results on urban land use discrimination may
be obtained by using a combination of co-occurrence tex-
ture measures. In particular, this procedure exploits the
spatial disposition of the man-made features, which have a
peculiar response in radar images. Co-occurrence features
highlight the spatial patterns of backscatterers. À super-
vised clustering of these features reveals where buildings
and other man-made objects gather in a are way. So, res-
idential areas with isolated scattering elements are quite
different from town centers with many crowded backscat-
terers or even financial areas. The methodology proposed
for exploiting these information consists of three subse-
quent steps: first, compute the co-occurrence matrix and
323
extract textural features, applying a priori knowledge, if
any, on the optimal scale or the best range of scales; sec-
ond, determine which feature set is the most useful to dis-
criminate the classes in the training set; third, classify the
chosen feature set using the same training areas as seeds
for a supervised clustering procedure. Classification maps
for two RADARSAT-1 images of the area of Pavia, North-
ern Italy, are shown in fig.1, and show the dependence of
the class accuracy to the incidence (beam) angle. It is in-
deed interesting to observe that the accuracy of the map
increases with more nadir-looking views, but this is due
mainly to the “water” and “sparse buildings” classes, while
the behavior of the areas where many strong scatterers are
present is less various.
5 CONCLUSIONS AND PERSPECTIVES
This work presented some of the recent efforts for the ex-
ploitation of EO data for the realization of GMES urban
service products. We highlighted for a couple of products
which are the methodologies available in technical litera-
ture and which are their weaknesses. Moreover, we intro-
duced an approach suitable for exploiting data from SAR
sensors, usually neglected in urban remote sensing appli-
cations.
The interest to urban products from EO data is increasing,
in parallel with the availability of more refined algorithms
for data interpretation. Moreover, the requirements by EU
and especially the UTS are driving the need for these infor-
mation, especially as aggregated indicators of urban qual-
ity and environmental characteristics. More work is there-
fore needed to integrate new data sources as well as to con-
nect more tightly the users with the producers via a suitable
application-oriented research and development effort.
ACKNOWLEDGEMENTS
The authors acknowledge the support ofthe European Space
Agency for this work, through the GUS project.
REFERENCES
S Barr and M. Barnsley, 2000. “Reducing structural clut-
ter in land cover classifications of high spatial resolu-
tion remotely-sensed images for urban land use mapping,”
Computers & Geosciences, vol. 26, pp. 433-449.
K.S. Chen, S.K. Yen, D.W Tsai, 1997. “Neural classifica-
tion of SPOT imagery through integration of intensity and
fractal information,” Int. J. Remote Sens., vol. 18, n. 4, pp.
763-783.
K. Chen, 2002. “An approach to linking remotely sensed
data and areal census data,” Int. J. Remote Sensing, Vol.
23, n. l, pp. 3748.
F. Dell’ Acqua, P. Gamba, 2003. “Texture-based character-
ization of urban environments on satellite SAR images,”
IEEE Trans. on Geoscience and Remote Sensing, Vol. 41,
n. 11, pp. "153-150