Accurate Coregistration of Mixed Scenes of Stationary and Non-Stationary Areas for DInSAR Applications
Riva, Davide; Giovannini, Luca; Giudici, Davide; Piantanida, Riccardo
Aresys srl, ITALY

SAR Interferometry is a radar remote sensing technique renowned for its applications to surface elevation map generation and surface deformation monitoring. More specifically, using the technique of Differential Interferometry (DInSAR) it is possible to identify the movements of non-stationary areas on the Earth's surface (such as landslides, glaciers, polar ice flows, etc) and to estimate their extent and amplitude. However, like all interferometric techniques, DInSAR is very sensitive to the problem of phase decorrelation. For surface movements this problem is most commonly due to displacements that are much greater than the image spatial sampling, which in turn is influenced by a combination of factors, mainly: the actual resolution of the images, the speed of the surface movements and the time separation between the different SAR acquisitions used. For instance, Sentinel-1, the forthcoming SAR satellite from ESA, due to its high resolution is more prone to this phase decorrelation problem, especially in fast-moving non-stationary areas such as polar ice flows.

In this work we describe a technique for the accurate coregistration of scenes with mixed stationary and non-stationary areas in high-resolution SAR images. The aim of this technique is to coregister images not only with respect to the effects of surface geometry (stationary coregistration component), but also to compensate for the effects of surface movements (non-stationary coregistration component). To do so, the technique applies an intensity cross-correlation approach to match the corresponding surface features in the images that works, as such, also on areas affected by phase decorrelation like non-stationary areas. Moreover, the non-stationary coregistration component is isolated from the global estimate and filtered to enhance movement features and to limit the amount of estimation noise, especially at the boundaries between stationary and non-stationary areas, whenever present. As a consequence of this procedure, images are globally coregistered to sub-pixel accuracy and can be employed for DInSAR analyses and related applications.

The technique has been tested on real and simulated data and the results show a relevant increase in phase correlation (i.e. higher coherence) after coregistration with respect to reference standard coregistration techniques. The results of these tests are presented and critically discussed.