Assessing closure phase statistics and its impact on InSAR time-series

Abstract:

Interferometric Synthetic Aperture Radar (InSAR) time-series analysis techniques usually assume that the interferometric phase is linear with displacement of the ground which implies zero sum of interferometric phases among three acquisitions. This underlying assumption is challenged by the presence of systematic non-zero closure phase, other known as phase inconsistencies, or non-zero phase triplets in multi-looked InSAR measurements.

Phase inconsistencies arise when we spatially average InSAR measurements. The phase of averaged InSAR observations is the Maximum Likelihood Estimate (MLE) of the interferometric phase of the area under the assumptions of independent observations and statistical homogeneity of the multi-looking area. The MLE of phase is unbiased when the number of samples tends to infinity. In this work, we investigate conditions that may lead to biased phase estimates, which in turn results in incomplete or inconsistent phase closure. Specifically, we examine data homogeneity, and the statistics of intensity and phase of single-look measurements inside the multi-looking window.

We examine the statistical properties of closure phase through different case studies and evaluate statistics of phase inconsistencies with respect to land cover type, correlation level and temporal baseline. The results demonstrate that the spatial distribution of closure phase is highly correlated with land-cover types and systematic non-zero phase closure of tens of degrees can be found in areas with interferometric correlation.

To assess uncertainties introduced by systematic non-zero closure phase in InSAR time-series, we compare InSAR time-series derived from different inversion methods including wrapped phase series estimated with MLE and Eigen-Value-Decomposition methods of coregistered stack of SLCs as well as least squares of stacks of unwrapped interferograms. For each method we investigate the impact of network connectivity (e.g., all possible pairs versus subsets of pairs) on the uncertainty of the estimated time-series. Finally, we discuss implications of systematic phase inconsistencies on designs of InSAR time-series algorithms and the uncertainty of the displacement time-series products.