Permanent scatterers in SAR interferometry

Differential SAR interferometry (DInSAR) is an unique tool for low-cost, large-coverage surface deformations monitoring. As well known, the technique involves interferometric phase comparison of SAR images gathered at different times and has the potential to provide millimetric accuracy. Though temporal decorrelation and atmospheric dishomogeneities strongly affect interferogram quality, reliable deformation measurements can be obtained in a multi-image framework on a small subset of image pixels, corresponding to stable areas. These points, hereafter called Permanent Scatterers (PS), can be used as a ”natural GPS network’’ to monitor terrain motion, analysing the phase hstory of each one. In this paper, results obtained using 34 ERS SAR images gathered over the Italian city of Ancona are presented. 1 Permanent Scatterers Despite the tremendous potential of DInSAR technique, to date accurate routine monitoring of landslides, volcanoes and seismic faults still remains a challenge. Limitations are essentially due to temporal decorrelation and atmospheric dishomogeneities [ 11. Temporal decorrelation makes InSAR measurements unfeasible over vegetated areas and, in general, where the electromagnetic profile and/or the position of the scatterers within each resolution cell changes with time. On the other hand, atmospheric dishomogeneities introduce low wavenumber artifacts in estimated DEMs and deformation patterns [2]. The main goal of our research is the identification of single coherent pixels, so called Permanent Scatterers (PS), starting from several SAR images separated by large baselines (even larger than the decorrelation one) in order to get sub-meter DEM accuracy and terrain motion in low coherent areas on a pixel-by-pixel basis’. It will be shown that even if no fringes can be seen generating single interferograms, reliable elevation and deformation measurements can be obtaided on this subset of image pixels. The mathematical framework for this kind of estimation is rather simple. Let us suppose that N + 1 ERS SAR ‘This work was partially supported by ESA-ESRIN. A detailed description of the technique will be submitted to IEEE Trans on Geosczence and Remote Sensang. images of the area of interest are available. Data are first coregistered on a unique master and a DEM of the area is estimated starting from low temporal baseline pairs [2 ] . Then, N differential interferograms between all SAR images and the master are computed. After DEM compensation, the residual phase of interferogram i is: where X is the system wavelength, Aai the atmospheric phase contribution, Ani the decorrelation noise, Etopo-i the phase contribution due to possible errors in the DEM (proportional to the normal baseline of each image Bn-i) and ArTi is the possible target motion in the direction of the satellite line-of-sight (LOS). Time series analysis of the first term in equation 1 should reflect target motion: here we will suppose a constantvelocity model (a uniform strain rate hypothesis is often used in geophysical modeling). Of course, more complex models can be adopted if the linear model is unrealistic. Therefore, the first term in equation 1 can be written as follows: where U , is the unknown component of the target velocity in the direction of LOS and Ti is the temporal baseline between the master acquisition and the generic i-th slave image. Since we have N differential interferograms of the same area with different temporal and geometric baselines, we finally write, for each pixel, a linear system of N equations and 2 unknowns ( E ~ , U T ) : where E% is the DEM error and C,i is proportional to B,-i. In order to reduce the effect of atmospheric patterns and compensate for baseline errors, the low frequency components should be previously removed from pi. The processing is then performed on a pixel-by-pixel basis and the final result is a velocity field of the area of interest together with an improved DEM. 0-7803-5207-6/99/$10.00 Q 1999 IEEE. 1528 2 Results over Ancona Landslide