Homogeneous pixel selection for distributed scatterers using multitemporal SAR data stacks

Abstract

Land deformation and topography estimation using Synthetic Aperture Radar (SAR) interferometry has gained more and more attention over the last years due to its high quality. Up to now, there have been many high resolution SAR satellites launched, such as TerraSAR-X, TanDEM-X, and COSMO-SkyMed. Multitemporal interferometric SAR (InSAR) techniques have become important quantitative geodetic tools to monitor deformation time series. The Persistent Scatterer Interferometry (PSI) technique utilizes coherent radar targets exhibiting high phase stability. The technique is reliable for monitoring deformation with millimeter accuracy. Persistent Scatterers (PSs) e.g. man-made structures, boulders, and outcrops are widely available over a city, however, they have less density in non-urban areas. As a result, Distributed Scatterers (DSs) corresponding to image pixels belonging to areas of moderate coherence, where many neighboring pixels share similar reflectivity values (as they belong to the same object), are used in order to increase the density of measurement points, such as in the SqueeSAR, Small BAseline Subset (SBAS) techniques etc. In order to improve the estimation accuracy of interferometric phase and coherence of DSs, multilooking is needed, where neighboring pixels are averaged together with the target pixel to reduce phase noise. Conventional boxcar kernel multilooking technique is based on the hypothesis of statistical homogeneity of the averaged pixels in a rectangular window surrounding the DS pixel. Nevertheless, as the size of the kernel increases, this hypothesis loses its validity. There can be loss of details as pixels arising from different statistical distributions are averaged together. Therefore, an adaptive multilooking technique is required to preserve the high resolution provided by modern satellites. This thesis aims to find reliable and robust methods to select homogeneous pixels, i.e. the pixels belonging to the same category of scatterers, to perform adaptive multilooking of DSs. The focus is on techniques that make use of the amplitude information of a multitemporal SAR data stack. Accordingly, two new methods, based on Confidence Interval (CI) of amplitude mean and CI of amplitude median have been developed as part of the thesis and principles behind these are clarified. Results of the selected methods are shown and compared using TerraSAR-X data from two different test regions (i.e. Lueneburg and Cologne). Performance, quality and robustness are analyzed. At last, the conclusions are summarized and some suggestions are proposed for future research. The algorithms can aid in accurate covariance matrix estimation and can be widely applied in topographic mapping and deformation monitoring.