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
ON THE USE OF DUAL-CO-POLARIZED TERRASAR-X DATA
FOR WETLAND MONITORING
A. Schmitt, T. Leichtle, M. Huber, A. Roth
German Aerospace Center (DLR), Earth Observation Center (EOC), Oberpfaffenhofen, D-82234 Wessling, Germany -
(andreas.schmitt, tobias.leichtle, martin.huber, achim.roth)(g)dlr.de
KEY WORDS: Environment, Land Cover, Dynamic, Change Detection, SAR, Algorithms, Multitemporal, Value-Added
ABSTRACT:
Today's SAR sensors provide a variety of different image modes particularly with regard to multipolarised acquisitions. Until now,
each polarisation mode requires a special decomposition which is a severe drawback when designing processing chains. Therefore, a
new description for multipolarized SAR data based on the well-known Kennaugh matrix was developed that enables the uniform
description and processing of SAR data independent of its polarisation by separating backscattering strength from polarimetric
information. This mathematical approach subsequently is extended to the processing of multitemporal SAR data in order to stabilize
the polarimetric information over longer periods of time and to enhance temporal changes in the polarimetric backscattering.
Because of the high sensitivity of the Kennaugh elements a novel multilooking technique based on the Gaussian pyramid is used that
locally adapts the look factor and thus selects the optimal balance between radiometric accuracy and geometric resolution. This
methodology is applied to two dual-co-polarized TerraSAR-X acquisitions over the RAMSAR testsite "Upper Rhine" in order to
generate value-added products that help to map land cover and land cover changes in consequence of water level changes. The first
results are very promising although the interpretation of the observed polarimetric changes is not yet validated. The aim of this paper
is to present a further application of the (Differential) Kennaugh matrix which will be the kernel of a polarimetry and change
detection processor to be implemented in the coming years.
1. INTRODUCTION
This paper addresses the use of TerraSAR-X data in order to
determine the extent and the spatial as well as the temporal
variability of open water and flooded vegetation surfaces in
wetlands.
1.1 Background
Previous studies use either single- or quad-polarized (Brisco,
2011) data for wetland mapping. Single-polarized data is
always available because it is still the standard mode of the
common space-borne SAR sensors. Open water surfaces can be
identified in SAR images by their low backscattering like it is
done generating the water indication mask of the TanDEM-X
mission (Wendleder, 2011). Flooded vegetation is expected to
show a strong double-bounce backscattering. Unfortunately, the
backscattering mechanism cannot be identified in single-
polarized data sets. Therefore, the extent of flooded vegetation
areas can only be estimated by their stronger backscattering in
each (co-pol) channel (Hess, 1990). Quad-pol data enables the
identification of several backscattering mechanisms. But, quad-
pol data is not easy to acquire. Most SAR satellite sensors do
not support the polarimetric mode at any user-defined time. As
the quad-pol mode always effects a reduction of the imaged
area and the resolution as well, this method is not suitable yet
for large-area applications. Hence, we focus on dual-co-
polarized data. The combination of the complex HH and VV
allows the identification of two scattering mechanisms: surface
and double-bounce. The reduction of image size and resolution
is lower than using quad-pol data. Additionally, dual-co-
polarized data is available at any time and in almost any mode
with TerraSAR-X and several other sensors.
1.2 Methodology
The two complex layers of the TerraSAR-X dual-polarized SSC
product cannot directly be interpreted and have to be
decomposed and geocoded. As all common polarimetric
decompositions only concern quad-polarized data, a new
Kennaugh matrix like decomposition has been developed
(Schmitt, 2012). The decomposition produces four layers. The
first comprises the total intensity of both layers, which is very
low over open water surfaces. The second layer gives the
intensity difference between surface and double-bounce
scattering. Thus, high values in the second layer indicate a
dominant double-bounce scattering while low values indicate a
dominant surface scattering. The third layer includes the co-pol
diattenuation which is represented by the intensity difference
between the two measured channels. The fourth layer is
negligible only holding complementary correlation information.
All layers are normalized by the total intensity or the so-called
Hyperbolic-Tangent-Normalization respectively so that their
value range is reduced to ]-1,1[. Having a closed range the
values can be saved in an integer format without clipping
information at the upper or lower end. The chosen bit depth
then only fixes the sampling which reaches its maximum near
the centre and decreases towards higher deviations in both
negative and positive direction. For display and interpretation
reasons the Hyperbolic-Tangent-Scaling of Kennaugh elements
can directly be transferred to the common unit decibel via the
Inverse Hyperbolic Tangent function. The first layer represents
the total intensity independent of the polarization state while the
other layers represent the polarized contribution to the total
intensity. Hence, intensity and polarimetric information can be
evaluated separately.
