InSAR uncertainty due to phase unwrapping errors

Abstract:

InSAR phase observations are wrapped and known only modulo 2pi. Recovering the unambiguous phase from the wrapped phase by adding integer numbers of 2pi, known as phase unwrapping, is a key step to obtain the continuous phase field for DEM generation and surface deformation mapping. Wrong integer numbers of 2pi added to the wrapped phase during phase unwrapping, to which we refer as unwrapping errors, could bias the height or displacement measurements. A better understanding of this error and its contribution to the uncertainty budget is crucial to quantify the uncertainty of InSAR displacement products. Moreover, a realistic InSAR uncertainty model that accounts for unwrapping errors is needed for the performance analysis of future SAR missions during design, development and operation.

We quantify the uncertainty of InSAR observations due to unwrapping errors using a statistical approach with data simulations. For this purpose, we simulate realistic interferometric phase contributions from various sources, including decorrelation using the observed coherence from C- (Sentinel-1) and L-band (ALOS-1) missions, dry and wet tropospheric delay using ERA5 and spatial spectra database from MODIS/MERIS, and long spatial wavelength contribution from ionosphere using GNSS-based Total Electron Content. We realize the simulation over multiple datasets with different noise characteristics, unwrap each realization with different algorithms including the Minimum-Cost-Flow and evaluate unwrapping errors in terms of likelihoods at pixels, regions and SAR scene levels.

Results of the simulation realizations in different regions demonstrate that connected components are reliable indicators of potential unwrapping errors with <2% false positive rate. Without significant phase gradients from fast ground deformation or from atmosphere, the magnitude of unwrapping errors is bounded within 8pi and is a function of distance between connected components. The occurrence probability of unwrapping errors from repeated realizations for a low-correlation Sentinel-1 scene over central San Andreas is shown below. We model the unwrapping errors as a function of radar frequency, scene correlation and topography and SAR geometry and extend the analysis to a global scale using an existing global correlation database.