7°30E 7*40' T°SCE
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
sous
q^ SWITZER|
i ;
+ A
d Sond E
Data Sources:
Basemap: Bing Maps Aerial
Cities & Towns: | www.openstreetmap.org.,
www.mapcruzin.com
National borders: EuroGeographics
(epp.eurostat.ec.europa.eu)
Coordinate System: UTM (WGS 1984) Zone 32N
Legend
® City
® Town
Nationat border
Study Area
GN GENS CEBSEENEEEEENNE Kilometers
0 2 4 8 12
Figure 1: Physical Map of the Study Area “Taubergieflen” in south-west Germany
A new multilooking approach based on the Gaussian pyramid
is used in order to suppress noise and to preserve the images
geometric resolution because multiscale approaches have
shown a high potential in previous studies (Schmitt, 2009).
Instead of filtering the image with a constant number of looks
this method locally adapts the number of looks to the image
content by selecting the suitable image scale out of the
Gaussian pyramid. Starting from the coarsest scale, i.e. the
largest number of looks and hence, the best radiometry, the
intensity values of the finer scales only are adopted if their
value deviates significantly from the value with a higher look
factor. The decision between better radiometric accuracy and
geometric resolution is made by evaluating a stochastic model
for the comparison of multilook intensities that is based on the
beta distribution. Large homogeneous areas are smoothed with
a large number of looks while fine structures showing
deterministic scatterers remain unsmoothed. The decision
pyramid derived from the intensity layer is subsequently
applied to all available polarimetric layers.
The resulting layers are geocoded via the TerraSAR-X
geocoding routines and normalized as described above. As this
algorithm is not restricted to single-pass acquisitions, it also can
be used to compare the polarimetric scattering behaviour
reported by several repeat-pass images. Two image acquisitions
can be merged by adopting both as independent looks for one
resulting “mean” image. Beside that they can be merged as well
by establishing the so-called differential Kennaugh matrix
(Schmitt, 2012) that comprehends the change in intensity and
polarimetry in separate layers similar to the Kennaugh matrix
for single pass acquisitions. Even the pyramidal multilooking
can be used to enhance the information content. The decision
pyramid is drawn from the total intensity of both images —
equivalent to the first laser of the “mean” image — and then
applied to all differential Kennaugh layers. Finally, the
normalization step converts the changes to the Hyperbolic-
Tangent-Scaling and enables further conversion to the usual
unit decibel.
In summary, a combined description of single- and multi-pass
acquisitions for all common polarimetric SAR data modes from
single- via dual- to quad-polarization was developed.
Additionally, the new multilooking approach guarantees robust
and reliable radiometric results without losing geometric
resolution. Consequently, simple thresholding whether by
empirically or statistically derived thresholds delivers clear and
smooth classification results. The final value-added product
displays clearly open water surfaces and flooded vegetation for
single-pass images and the change in open water and flooded
vegetation for repeat-pass images.
1.3 Testsite
As test area the RAMSAR site “Upper Rhine” in southern
Germany is chosen which is a wooded flood plain near the river
Rhine, see Fig. 1. Because of its very small extension and its
fine structures high resolution SAR data is absolutely required.
The test site covers large areas of the nature reserve
"TaubergieBen", which is an important bird area and thus
belongs to the Natura 2000 network. The wetland itself is
dominated by freshwater swamp forests and wooded swamps,