Titre/Title :

Imagerie Radar à Synthèse d’Ouverture interférométrique et polarimétrique. Application au suivi des glaciers alpins / Interferometric and Polarimetric SAR Image Processing. Application to Alpine Glacier Monitoring


Abstract :

Synthetic aperture radar (SAR) imagery is a powerful information source for monitoring terrestrial geophysical objects by remote sensing techniques. In this study, two main axes were investigated. Firstly, the methodology of quantitative remote sensing is discussed in the context of multivariate polarimetric (POL) or/and interferometric (In) SAR images. Secondly, the limits and the potential of the POL/In/POL-InSAR imaging techniques are evaluated for monitoring temperate glacier from the “Chamonix Mont-Blanc” test site (France).

After a brief state of the art in the field of polarimetric and interferometric SAR image processing and analysis, the first methodological part of this thesis proposes a particular focus on speckle filtering of POL/In/POL-InSAR multivariate data and its implication on parameter estimation. The objective of speckle filtering is to retrieve the radiometric and spatial scene information from the observed “speckled” SAR measurement. The general formalism used for data representation is given by the second order statistical descriptors, namely the coherency matrices. Up to now, the majority of conventional speckle filters are based on neighbourhood operators. Such operators require local hypotheses of stationary and ergodicity inside the estimation neighbourhood. Three main directions for gaining in stationarity/ergodicity have been investigated: signal adaptive neighbourhoods (boxcar, directional, region growing), non-adaptive/adaptive estimators (complex multilooking, LLMMSE) and compensation of deterministic components in the interferometric phase signal (InSAR local frequencies). Also, several consequences of speckle filtering are discussed in terms of robust parameter estimation (e.g. coherence maps, phase unwrapping, Cloude and Pottier polarimetric decomposition or POL-InSAR coherence optimisation).

This thesis proposes a novel strategy for filtering multivariate SAR images, namely the use of adaptive neighbourhoods obtained by multivariate region growing techniques. The principle consists in associating to each pixel of the image a neighbourhood whose shape and size are adapted to the statistics of the local population to which this pixel belong to. According to this algorithm, named IDAN (Intensity-Driven-Adaptive-Neighbourhood), all the available intensity images of the polarimetric or/and interferometric components are fused in the region growing process to ensure the validity of the stationarity/ergodicity assumptions. Based on IDAN, a novel method for estimating local frequencies in SAR interferograms is proposed. Basically, the 2D phase signal is considered to have two deterministic components corresponding to low-resolution fringes and high-resolution patterns due to the local micro-relief, respectively. The first step of the proposed algorithm consists in the low-resolution phase flattening. In the second step the local high-resolution frequencies are estimated from the phase 2D auto-correlation function computed on adaptive neighbourhoods.

These algorithms have been tested both on several high resolution (metric) airborne data sets (at different frequency bands and polarisation configurations), and on low resolution (decametric) spaceborne tandem ERS SAR images. The obtained result recommend these techniques as possible alternatives for the processing of the future high resolution spaceborne data provided by recently launched or future SAR remote sensing satellites (TerraSAR-X, ALOS, RADARSAT-2, COSMO-SKYMED).

The second part of this thesis is dedicated to Alpine glacier monitoring by SAR remote sensing. This part tackles the problem of processing multivariate SAR data over the “Chamonix Mont-Blanc” test site. Although standard processing chains have already been applied for geophysical applications, the particularities of the proposed test site are quite challenging for SAR applications. Indeed, an elevation difference of almost 4000 m ASL between the Chamonix valley and the Mont-Blanc summit, coupled with a mean glacier velocity of about 30 cm/day, make standard SAR data processing quite difficult over this area. By using the specific algorithms presented in the methodological part of this manuscript and other approaches developed through several collaborations with various national (IETR – POLSARpro software or Telecom PARIS – SYTER processor) and international (JPL – ROIPAC processor) laboratories, the first analysis of the limits and the potential of monitoring Alpine glaciers by POL/In/POL-InSAR techniques is proposed. The first application presented consists in measuring the 3D displacement field of an Alpine glacier by differential SAR interferometry (D-InSAR). The analysis of several tandem ERS interferograms between July 1995 and April 1996 shows that it is possible to measure temperate glacier surface velocity fields from October to April in 1-day C-band interferograms with approximately 20-meter ground sampling. Also, the first D-InSAR velocity field over a glacier in the French Alps is presented.

The second application is represented by the first POL/POL-InSAR analysis of multiband high resolution airborne SAR data acquired by the DLR E-SAR system in October 2006. Several glaciological applications which could benefit from the surveillance by polarimetric/interferometric SAR remote sensing are finally identified (characterisation of the snow cover or the Forbes bands phenomenon).