POLInSAR Workshop 2003
Topic : Polarimetry/Interferometry
Chair/s S Cloude, D Corr, H Skriver, J C Souyris
There were eight papers presented over two sessions in this topic area.
Included were two overview papers, summarising the findings of the two
consortia funded under the ESA-ESRIN project “Applications of SAR Polarimetry”.
There were two further papers concerned with surface parameter estimation
using full polarimetry (surface slope estimation) and polarimetric interferometry
(surface roughness and moisture content). Two papers were then presented
on the implications of employing full polarimetry and interferometry for
space and airborne system design. One paper was concerned with polarimetric
radar classification of a mountainous natural forested area and a comparison
with optical classification results. Finally a paper was presented on the
use of sub-aperture techniques to study the variation of full polarimetric
response across the synthetic aperture.
Conclusions and recommendations.
Arising from these presentations and the round-table discussions, we
have identified three important topic areas that we suggest form the basis
for future studies in the area of POLSAR and its applications:
Topic 1 : The Importance of Full versus Partial Polarimetry in Radar
Strong evidence was presented from both consortia that the use of fully
polarimetric data provides improved discrimination and classification results
for land, sea ice and forestry applications. In addition, fully polarimetric
data also provides improved quantitative parameter estimation in surface
roughness and moisture studies. The importance of obtaining both polarimetric
phase and amplitude information were once again highlighted. This can be
expressed as a recognition that there is important information contained
along the diagonal of the coherent scattering matrix and not just in one
column. Hence a major recommendation of this group is for ESA to continue
developments towards fully polarimetric space based sensors.
The question of the best radar centre frequency was also addressed.
From the results presented it seemed that L-band provides the best general
performance across all terrain types and has less class variability than
C-band. There were a few contrary observations demonstrating that for some
agricultural applications C-band shows better performance than L-band.
With the planned future deployment of a fully polarimetric C-Band sensor
(Radarsat 2) already at an advanced stage, it is recommended that ESA set
priority in providing a space borne fully polarimetric capability at L-band
for land, ice and forestry applications.
Topic 2 : The Implications of Full Polarimetry for Radar System Design
It follows from topic 1 that current partial or ‘light’ polarimetry
sensors such as ASAR are not optimally matched to the information content
of the polarimetric scattering matrix. Two solutions can be suggested to
improve this situation in future sensors. Full quadpol scattering matrix
measurements provide one approach and yield the widest range of applications
and best accuracy in classification and surface parameter retrieval (and
also for calibration issues related to Faraday rotation and system channel
cross-talk and amplitude/imbalance correction in the data). However the
required increased PRF causes swath coverage limitations, which are critical
for certain applications such as sea ice. Hence there is some trade-off
between coverage and information extraction in utilisation of this mode.
However, a new idea was presented at the workshop, which could provide
an alternative strategy for the design of polarimetric sensors. The p/4
mode involves a mixed basis measurement, whereby the polarisation of the
transmitter is constant but different to that of the orthogonally polarised
receiver. This then maintains lower PRF operation and hence the maximum
coverage of the sensor. However, because of the mismatch between transmit
and receive basis, there is a mixing of information along the diagonal
of the scattering matrix and hence there is improved classification and
parameter estimation compared with existing column based switching methods.
A detailed analysis of the covariance matrix of this mode was presented
and it was shown how extra assumptions must also be used to enable good
parameter estimation in this mode. It is not then well suited to point
targets but for distributed targets such as surface and volume remote sensing
the performance seems good.
It is recommended that further studies be made of this mixed basis mode
of operation to further assess the ability to extract surface and volume
parameters and to examine more closely the calibration and system cost
and design implications.
Topic 3 : Techniques for Classification and Segmentation of POLSAR Images
It was recognised by both consortia that the performance of Maximum
Likelihood Wishart based classifiers was sometimes inferior to other methods.
This was attributed to the fact that such classifiers employ the full covariance
matrix and hence, in addition to information channels, contains some noisy
channels, which reduce the reliability compared to some other methods.
It was recognised that, while the Wishart approach is statistically optimum
in some sense, hybrid methods employing sub-space parameter methods can
provide improved performance. Several of these were presented at the workshop,
including hierarchical adaptive threshold methods and log-normal intensity
channel methods. The last of these showed particularly good results, although
only on one data set, and it is recommended that the method be tested on
other data sets before deciding on its suitability for recommendation as
a good strategy for general classification studies.
The performance of supervised and unsupervised Wishart classifiers will
be improved by applying constraints based upon phase characteristics of
the covariance data. A validated inter-cluster distance measurement for
cluster management is also required.
It was further noted that there now exist a wide range of polarimetric
parameters for investigation across the spectrum of applications. To assist
users in the selection of parameters for their applications, a set of software
tools are being developed. One such tool set was presented by one of the
consortia and is to be made available as an output from the workshop. It
is recommended that further such tool sets be developed and provided to
the user community.
One final topic of interest was the study of statistical fluctuations
in the data. There are two main sources of these, speckle and natural class
variability. Techniques for handling speckle have been widely considered
and there is now a wide choice of mature methods available to the user.
However, the studies showed that class variability can sometimes be the
dominant source of fluctuations in the data. However this term tends to
be frequency dependent with higher variability at C than L bands. This
again lends support to the idea of deployment of a future L-band quad-pol
A Review of the
Applications of SAR Polarimetry and Polarimetric Interferometry - an ESA-funded
Dr Douglas Corr
Synthetic Aperture Radar Polarimetry
Dr Henning Skriver
A Review of Polarization
Orientation Estimation from Polarimetric SAR Data
Dr. Jong-Sen Lee
Estimation Using Interferometric Cohrences between Different Polarisations
What can be learnt
from a half-polarimetric SAR?
Dr Jean-Claude Souyris
Why We Do Need to
Place Multi-band Single and Multiple Pass POLinSAR Monitoring Platforms
Professor Wolfgang-Martin Boerner
interferometry classification, resolution estimation and comparison with
optical images in Glenaffric radar projrct.
Dr. Parivash Lumsdon
Analysis of SAR
Response Anisotropic Behavior Using Sub-Aperture Polarimetric Data
Dr. Laurent Ferro-Famil