Information extraction: from SAR data to ground displacement measurements

Person in charge: Marie-Pierre Doin, LG-ENS

Natural hazards assessment relies on the accurate detection of ground displacements. SAR data are potentially extremely powerful to map ground deformations with a dense spatial coverage and a regular temporal sampling. However, they have been so far under-exploited for operational applications, in favour of "more reliable" but expensive in situ local measurements. Some methodological developments should contribute to overcome some of the limitations of radar-based techniques for soil deformation mapping (this will rely on the results of WP1 and WP2). The methods developped in this second subprroject are dedicated to measure small (mm to tens of cm) and large (> 0.5 m) displacements and based, respectively, on the use of the radar coherent return (radar interferometry) or amplitude image correlation.

Small displacements detection

Exploiting a huge amount of satellite data (ERS, Radarsat, Envisat, ALOS, ...), we aim at characterizing ground deformations of different origins in various environments. In urban setting, deformations can be due to changes of water level or content in aquifers, swelling clays, and can vary with meteorological events. In "natural" setting, surface deformations are related to faults and volcanoes activity, landslides, loading of anthropic and natural origin (glacier retreat, mining, or lake impoundment). The "end-member" users of deformation maps who contribute to assess natural hazards in these contexts specifically need to:

* detect ground motion with a few cm to mm accuracy,
* map deformation on spatial scales of up to 100 km,
* measure transient deformations events, non linear in time.

Radar interferometry has recently proven to be able to meet all these requirements with a high success [Cavalié et al. 2007, Doubre et al., 2007, Usai et al., 2003], when a large number of multi-baseline radar acquisitions at the same location and with close viewing angles (number of images > 30, hereafter called "data pile") are merged together. A large data pile is necessary to overcome the influence of dominant atmospheric artefacts (equivalent to up to ~6 cm of displacement in the radar line of sight). However, until now, such studies have been successfull only in desert, non mountainous, areas, where coherence is high. Coherence loss, associated with varying viewing angle and changes in surface backscatter properties, is therefore today the main stone walk towards systematic InSAR exploitation for small ground motion measurements. In this subproject, we intend to improve the extraction of interferometric phase informations in areas of coherence loss, relying on the exploitation of a large data pile.

Large Displacements detection

In the case of large displacements, the coregistration between radar acquisitions must take into account ground deformation, before exploiting the phase differences measurements. For a glacier, as the velocity varies from the borders to the center on a relatively short distance (~500 m), the coregistration step is more complex and should allow large deformations over small distances, which is not taken into account in classical software. When phase information is lost (due to coherence loss, unwrapping difficulties, as for the urban subsidence in Mexico city), large ground motions can only be monitored by a local registration step, which must then be as accurate as possible. Moreover, when polarimetric data are available, this local registration can take into account optimal polarization states or incoherent correlation techniques as developed in the Work-package 2.