ANALYSIS OF X-BAND POLARIMETRIC SAR DATA FOR THE DERIVATION OF THE
SURFACE ROUGHNESS OVER BARE AGRICULTURAL FIELDS
. . . , 3 ~ | fo 53 A a o ELS r uod
Nicolas Baghdadi', Noha Holah!, Pascale Dubois-Fernandez?, Laurent Prévot', Steven Hosford , André Chanzy', Xavier Dupuis”, Mehrez Zribi
! French Geological Survey (BRGM), 3 avenue C. Guillemin, B.P. 6009, 45060 Orléans cedex 2, France, n.baghdadi@brgm.fr
? ONERA, Base Aérienne 701, 13661 Salon AIR cedex, France
3 INRA, Unité Climat Sol et Environnement, Domaine St Paul, 84914 Avignon cedex 9, France
* CETP/CNRS, 10/12, avenue de l'Europe, 78140 Velizy, France
Commission PS, WG 1/3
KEY WORDS: Remote Sensing, SAR, Soil, Land Cover, Land Use, Agriculture
ABSTRACT:
We investigated the potential of fully polarimetric data in X-band to discriminate the principal surface types present in a study site
near Avignon in France: bare soils, agricultural fields, orchards (various fruits), forest, buildings, residential houses, and roads.
Decomposition and analysis techniques have been applied to a data set acquired by the ONERA airborne RAMSES SAR. The
discrimination potential of various polarimetric parameters (entropy, a -angle, anisotropy) have been discussed. Results show that
X-band provides some discrimination capability. The polarimetric parameters, entropy and « -angle, show clearly that these
classes’ signatures are grouped in five clusters corresponding to physical scattering characteristics. The introduction of the parameter
anisotropy does not improve our ability to distinguish between different classes whose clusters are in the same entropy/ @ -angle
zone. We observe a very weak correlation between the signal radar and the surface roughness over bare soils.
1. INTRODUCTION
A polarimetric SAR system measures the complete complex
scattering matrix [S] of a medium with quad polarizations. This
matrix is made up of the complex scattering coefficients Spy
where p is the transmitting polarization and q the receiving
polarization (p , q = H (Horizontal) or V (Vertical)). The
polarimetric information of a target can be represented by a
scattering coherency matrix [T] which can be calculated
directly from the complex scattering vector as follows (Cloude
and Pottier, 1996):
(7) =(kp 45")
with
SHH+SVV
Ets SHH -Syy
2SHV
where k, is the target vector of [S], the superscripts "T and
<> denote respectively the complex conjugate, the matrix
transpose, and spatial averaging over a group of neighbouring
pixels using a sliding window. Coherency matrices are
frequently processed for speckle reduction by averaging the
pixels in a 5x5 window.
Landcover classification using a fully polarimetric SAR image
is one of the most important applications of radar polarimetry in
remote sensing. Over the last few years, polarimetric data has
been successfully used for various applications for example: the
classification of sea ice, the derivation of land surface
parameters (surface roughness and soil moisture), and the
classification of trees according to age in a forest environment
(Scheuchl et al., 2001; Hajnsek et al., 1999; Ferro-Famil and
Pottier, 2001; etc.). However, the majority of these studies have
been conducted using polarimetric data in P, L and C bands. In
this paper, we have applied polarimetric processing techniques
to SAR data in order to evaluate the potential of X-band. The
overall objective of this study is to evaluate fully polarimetric
information by analysing the separability between the different
landcover classes present on the study site and thus derive the
parameters that are of most use in their classification.
2. POLARIMETRIC SAR DATA DECOMPOSITION
Cloude and Pottier (1996) have proposed a polarimetric
decomposition theorem based on the eigenvalue/eigenvector
decomposition of the coherency matrix into elementary
mechanisms (i.e. single, double and volume scattering) in order
to identify the global mean scattering mechanism. The matrix
[T] can be decomposed into its eigenvector basis using the
following equation:
3 s
(r)- X vw
i
where À; are the three eigenvalues of «[T]», real and non-
negative À4;22522420. V; are the related orthogonal unitary
eigenvectors. The eigenvectors are parameterised using 4
angular variables leading to an interpretation of the scattering
phenomenon:
V; 2 (cosa; sina; cos f; e^ ^ sine; sin f; on]
where a; , B; , 6; , and y; represent a set of four independent
parameters characterising the fully polarised backscattered
field.
Using eigenvectors and eigenvalues, three main parameters are
used to characterise the results of this decomposition: entropy
(H), @ -angle, and anisotropy (A). The entropy H is defined
from the logarithmic sum of eigenvalues of <[T]> and
represents the random behaviour of the global scattering:
International
where P; are t!
The entropy
mechanisms.
mechanism (
indicates a t«
with equal prc
The paramet:
mechanism
eigenvalues o
where a; are t
eigenvectors.
volume scatte
metallic surfa
The anisotro
significant eis
where A, and
It becomes (
proportion wl
anisotropic sc
Fully polarir
airborne RA
Center (ONE
Avignon (so
width and 5
ranging from
the range be
were apparen
ONERA. The
were 0.66 ar
image used 1
polarization ©
HV as green,
The study s
includes agri
roads. The ag
and wheat fi
among them
the image is
buildings ca
houses, low
more). In ge
roofs (mostl
survey, four |