Accurate deformation isolation in MTInSAR: parameter modeling and estimation

Abstract

Spaceborne interferometric synthetic aperture radar (InSAR) has proven to be a powerful tool for mapping ground deformation with large coverage and high spatial resolution. By exploiting the phase difference from two temporally and/or spatially separated SAR images of the same area, InSAR technology provides ground displacement information with centimeter to millimeter level accuracy. In addition to the deformation measurement, the interferometric phases contain multiple error sources, including topographic residuals, atmospheric delays, and decorrelation noise, etc. There is no doubt that reliable phase measurements are the basis of accurate deformation retrieval. However, conventional differential InSAR (DInSAR) cannot effectively handle the error separation. To overcome the limitations of conventional DInSAR, InSAR technique has evolved to leverage a stack of SAR images to improve the quality of deformation estimation. The developed technique is referred to as multitemporal InSAR (MTInSAR). After approximately 20 years of development, three categories of MTInSAR techniques are currently in existence. The first category of methods relies on exploiting single-master interferograms from the scatterers that have strong phase stabilities over a long period (i.e., persistent scatterers (PSs)). The second category of methods attempts to extract deformation information from the scatterers with moderate phase stabilities (i.e., distributed scatterers (DSs)), which usually appear in interferograms with short baselines. The third category is a hybrid of the first and the second approaches that makes use of all interferograms to enhance the phase quality of DSs, and then uses the combination of PSs and DSs to retrieve deformation informtation. Benefiting from the multitemporal analysis, the current MTInSAR techniques could mitigate the unexpected error sources in some degree. However, as various signals with different spatiotemporal characteristics are mixed into the interferometric phases, it is still challenging for current MTInSAR approaches to accurately isolate the deformation signal from the confounding measurements. Consequently, the derived displacements contain large uncertainties, which further reduce the applicability of InSAR technique in deformation mapping. We first propose an arc-based method to remove the decorrelation noise in multitemporal interferogram stack. The proposed method reconstructs interval phase maps based on the analysis of complex coherence information. The main feature of the method is the iterative weight update that counteracts the violation from phase ambiguity and coherence estimation deviation. In addition, linear formation makes the implementation suitable with small subset of interferograms, providing an efficient solution for processing big SAR data in future. Secondly, considering that the topographic error has a fixed spatial pattern and its phase contribution is baseline-dependent, we propose a non-parametric method to estimate the topographic error by using independent component analysis (ICA) decomposition. The ICA-based estimation does not need a priori information about the deformation during topographic error separation. Experiments from both simulated and real datasets show that the method can provide a robust estimation despite the phase observations are affected by atmospheric delays and/or the number of interferograms used is limited. Complementary to that, further progress is made to correct height-dependent tropospheric delays in MTInSAR analysis. The method simultaneously estimates stratified tropospheric delays with the parameters of deformation rate and topographic error based on their distinct spatiotemporal characteristics. The spatial variability of the tropospheric delays is addressed through localized estimation in windows which are derived by quadtree segmentation according to height gradient. The applicability of the three methods described in the thesis is examined using both simulated and real experiments. We integrate the three methods in MTInSAR framework and apply them to retrieve the slope movement before the collapse at Freeway No. 3 in northern Taiwan. The deformation results confirm the effectiveness of the error source correction and reveal that the downslope movements are correlated with local precipitation.