Non-linear Phase Linking using joined Distributed and Persistent Scatterers

We describe a python package for nonlinear phase linking of full resolution SAR images using both distributed and persistent scatterers. In the workflow, the first step is to find for each pixel the set of self-similar pixels in order to identify persistent and distributed scatterers. Next the phase linking is performed using the full complex coherence matrix containing the wrapped phase values of each distributed scatterer. We perform a combined eigenvalue maximum likelihood phase linking, which switches to the classic eigenvalue decomposition method for pixels with a non-invertible covariance matrix. The approach includes sequential phase linking. We then unwrap the phase by selecting an optimum unwrapping network of interferograms and invert for the unwrapped phase time series which is converted to the displacement time series. We show how the performance of phase linking depends on the temporal correlation behavior using simulations of the coherence matrix. The sequential approaches better retrieve the simulated phases compared to the non-sequential approaches for all temporal coherence models. Phase linking methods retrieve the simulated phase with residuals close to the Cramér–Rao lower bound for coherent seasons where the absolute values of coherence matrix are high and provide a tool for obtaining InSAR measurements over areas with seasonal snowfall. We furthermore show that unwrapping errors propagate differently depending on the unwrapping network. For single reference networks there is no error propagation, but for sequential networks it compromises the accuracy of the final displacement time series. Delaunay networks provide an optimum solution in terms of accuracy and precision if there is several years of data with frequent temporal decorrelation or strong seasonal decorrelation. We present applications using Sentinel-1 data in different natural and anthropogenic environments.