Cheap, Valid Regularizers for Improved Interferometric Phase Linking
Simon Zwieback
Abstract
Retrieving a consistent phase history from a multi-looked InSAR stack depends critically on the accuracy of the (coherence) magnitude estimates. To estimate the magnitudes more reliably, I propose three regularization methods: Hadamard, spectral, and Hadamard–spectral regularization. All three are computationally cheap and parameterized such that they are guaranteed to yield valid magnitude matrices. These regularizers achieved relative improvements in the phase history accuracy of up to 40% in simulations. The improvements were greatest for low long-term coherences. All three methods performed similarly in the simulations and in a Sentinel-1 stack, for which the local phase dispersion decreased with regularization. Implementation of the regularizers into operational processing chains is expected to improve deformation and uncertainty estimates, especially for local movements over decorrelating terrain.