Single-Network Wide-Area Persistent Scatterer Interferometry: Algorithms with Application to Sentinel-1 InSAR Data

Tuesday, 15 December 2015
Poster Hall (Moscone South)
Kanika Goel, Robert Shau and Nico Adam, German Aerospace Center DLR Oberpfaffenhofen, Oberpfaffenhofen, Germany
Advanced InSAR techniques, for example, Persistent Scatterer Interferometry (PSI), allow long term deformation time series analysis with millimeter accuracy. ESA’s Sentinel-1 SAR mission employs the TOPS mode as the standard mode for acquiring InSAR data. It provides a continuous and large coverage at conventional resolution. The idea is to have a wide-area PSI for mapping countries and continents. Although PSI has been successfully demonstrated and validated in the past for various applications, there are some limitations for processing a large-scale dataset. First, PSI is most effective in urban areas which have a large number of stable scatterers. For large-scale PSI, even non-urban areas need to be processed; and this requires robust algorithms for scatterer selection, network construction and inversion, and atmospheric phase removal. Second, the computational load can be very high, due to which, the processing is usually divided into overlapping blocks and merged later. This can however lead to spatial error propagation. This paper presents algorithms which have been developed for a robust PSI reference network estimation, while mitigating error propagation. Instead of dividing the scene into overlapping blocks, a single network (i.e. arcs connecting the scatterers) is created for the full scene. The relative deformation and residual DEM are estimated for the arcs using the LAMBDA estimator. The relative measurements of the network are finally integrated via least-squares inversion. Here, the sparsity of the system of linear equations is exploited to deal with big data (e.g. 10,000,000 arcs for 500,000 scatterers is a typical configuration for Sentinel-1). A QR or LU parallelizable solver is used for fast inversion. Also, variances of the estimates are calculated using a selected parallel inversion method based on LDL decomposition. Demonstration of the algorithms for large-scale deformation monitoring is provided using available Sentinel-1 data for Germany.