DECOMPOSITION OF RADAR POLARISATION SIGNATURES FROM BUILT AND 
NATURAL TARGETS* 
Y. Dong (Research Associate), B. C. Forster (Professor) and C. Ticehurst (PhD Candidate) 
School of Geomatic Engineering, The University of New South Wales 
Sydney NSW 2052, Australia 
Telephone: + 61 2 385 4177, Facsimile: + 61 2 313 7493 
Email: yunhan@fatboy.geog.unsw.edu.au 
Commission VII, Working Group 1 
KEY WORD: SAR, Modelling, Intepretation. 
ABSTRACT 
A method of optimal decomposition of radar polarisation signatures is developed. In the model the backscat- 
tering consists of single (odd), double, Bragg and cross backscattering components, and the Mueller matrix 
is the sum of the Mueller matrices of these four scattering mechanisms. The technique of Weighted Least 
Squares is then used to find the optimal combination of these four components. Using NASA/JPL AirSAR 
data, the results of the decomposition are compared with built and natural targets and found to agree with 
the general understanding of radar backscatter. The distinct decomposition of polarisation signatures can be 
used as a basis for land use classification. 
1 INTRODUCTION 
Quad-polarised SAR (Synthetic Aperture Radar) 
data such as that acquired and produced by 
NASA/JPL AirSAR system, give more information 
about the ground targets than single polarised SAR 
data. Excellent work has been done on modelling 
and classification of polarimetric radar backscatter for 
both built and natural targets (Freeman and Durden, 
1992, van Zyl, 1989, Pierce et al., 1994, Cloude, 1991, 
Kwok et al., 1994, Zebker and van Zyl, 1991, Dur- 
den et al., 1991, Evans et al., 1988, and Ulaby and 
Elachi, 1990). Freeman and Durden (1992) present 
a three-component model in which three scattering 
mechanisms, i.e., volume scattering, double bounce 
scattering and surface scattering are included. "Che 
contribution of each mechanism is then estimated by 
solving four equations with five unknowns using the 
elements of the measured Mueller matrix. One of the 
unknowns 1s assigned a constant in order to solve the 
equations. À technique is given by van Zyl (1989) for 
unsupervised classification of scattering behaviour by 
selecting the dominant scattering mechanism. Each 
pixel is classified as either an odd number of reflec- 
tions (small phase difference between HH and VV), 
even number of reflections (larger phase difference be- 
tween HH and VV), or diffuse scattering (little corre- 
lation between HH and VV) by analyzing the elements 
of the measured Mueller matrix. Using image infor- 
  
*This work was supported by the Australian Research 
Council. 
mation such as co- and cross-polarised backscattering 
coefficients and the phase difference between HH and 
VV responses available in polarimetric data, Pierce et 
al. (1994) build a knowledge-based classifier. Pixels 
are then classified into four categories: tall vegetation, 
short vegetation, urban and bare soil. 
The observed radar polarisation signatures are not 
very similar to the constructed polarisation signa- 
tures developed using only the one of simple scatter- 
ing mechanisms. It reveals that the observed radar 
response is a combination of responses from various 
mechanisms. Since different scattering mechanisms 
give different polarisation signatures, they could be 
extracted from the measured Mueller matrix. This 
paper proposes an approach to decompose the po- 
larisation signature observed by radar into a combi 
nation of scattering contributions by four basic scat- 
tering mechanisms. These four basic mechanisms are 
(1) double bounce scattering, (2) Bragg scattering, (3) 
single (odd) bounce scattering and (4) cross scatter- 
ing, which form six equations with four unknowns plus 
four boundary conditions. The WLS (Weighted Least 
Squares) technique is used to find the optimal solu- 
tions. The approach is applied to different categories 
of images including forest, farmland, ocean and urban 
areas, and the results are discussed. The reconstructed 
polarisation signatures are seen to coincide very well 
with the observed signatures. It seems that a bet- 
ter understanding of these scattering mechanisms will 
assist in classification of land cover types from radar 
image data. 
196 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996