Abstract :
[en] Synthetic Aperture Radar (SAR) Remote Sensing already proved as an ideal solution to determine surface displacements, thanks to its day-and-night and cloud-free characteristics. Furthermore, more and more acquisitions are becoming available for users (Radarsat Constellation Mission, Cosmo Skymed, SAOCOM, Sentinel-1, and so forth).
To detect displacements, SAR has several techniques. Using SAR images at different times from slightly different points of view, we can observe surface movements by phase shift measurements between acquisitions. These techniques belong to the branch of differential interferometry (DInSAR, SBAS, MSBAS, PSI, MAI, BOI, and so forth). DInSAR can determine displacements according to the line of sight of the sensor with an accuracy that can go below the centimeter, with a sensor at several hundreds of kilometers distance. This partly explained the success of DInSAR. Then, to reconstruct the bi- or tri-dimensional displacements, we need other measures from other orbits. Combining a great number of images from several orbits, we can reconstruct the full vectorial components of the surface movement.
Unfortunately, this abundance of orbits is far from achievable everywhere on Earth. In particular, Antarctica has many geographical areas where only a limited number of orbits is available. Besides, techniques based on SAR interferometry are limited by other factors. Among them, the magnitude of the displacements can introduce a decorrelation such that the wavefronts combination emitted from two different times does not give a coherent signal. This temporal decorrelation is particularly remarkable in coastal regions of Antarctica, where the revisit time of Sentinel-1 (6 or 12 days, depending on the region) allows the scatterers to move from one picture element to another.
In these cases, it is possible to employ another family of techniques, based on the tracking of feature elements at the surface. In SAR remote sensing, we talk of speckle tracking. In speckle tracking, the technique uses two SAR images at different acquisition times. According to a defined spatial sampling, we search in the second image a translation in picture elements that maximizes the local correlation. From this translation is deduced a bi-dimensional displacement and, in fine, a velocity. This technique is less precise than InSAR-based methods but is less impacted by temporal decorrelation, while also directly brings a 2D velocity vector.
The limits of applicability between InSAR and Speckle tracking are not fixed and, when the two options are possible, we would always opt for phase-based measurements thanks to their incredible accuracy. It is in this context that coherence tracking was born by taking the best of the two approaches. Coherence tracking determines bidimensional displacements by maximizing the quality of the interferogram at a local scale, through the coherence estimation. The use of the phase in a tracking approach allows recovering
the location of ground scatterers in the second image. Then, it is possible to determine a tracked interferogram, that contains the displacement along the line-of-sight, with good accuracy. Coherence tracking is one way to circumvent the issue of temporal decorrelation induced by fast-moving areas.
Nevertheless, the TOPSAR acquisition mode introduces a phase bias to be taken into account before the processing. By steering its sensor during the acquisition, Sentinel-1 contains in its signal a strong azimuthal phase ramp. While this phase ramp can be canceled out in classical interferometry, this is not the case in a tracked interferogram.
In this research, we present the coherence tracking technique and the added-value of the phase information in offset tracking methods. Then we explain how to adapt the approach in TOPSAR data, in particular with Sentinel-1. Results are related to the study of Ice Shelves in East Antarctica. More precisely, we are focusing on the Roi Baudouin Ice Shelf, in Dronning Maud Land. Results are finally compared to traditional approaches.
