Robust Refinement Method for Coherence Matrix in Time-Series InSAR Phase Linking

Objectives Accurately retrieving time-series consistent phase information from time-series stacked data through phase linking (PL) serves as the critical processing step in hybrid time-series interferometric synthetic aperture radar analysis that jointly utilizes permanent scatterers and distributed scatterers. PL techniques based on maximum likelihood estimation are constrained by the estimation bias of the sample coherence matrix (SCM). During matrix inversion, this bias can be significantly amplified, potentially leading to the overemphasis of unreliable phase observations.

Methods To address the issue that non-Gaussian scattering vector perturbations and temporal decorrelation can introduce significant bias into the sample coherence matrix (SCM), this paper proposes the temporal decoherence taper shrinkage for SCM (TDTS-SCM) method. By leveraging both the statistical characteristics of the scattering vectors and the internal structure features of the initial SCM, the method achieves a bidirectional coupling between statistical properties of complex signals and underlying physical coherence mechanisms. Integrated with the eigendecomposition-based maximum likelihood estimator, the proposed approach enables accurate estimation of the SCM and robust recovery of time-seires consistent interferometric phase.

Results Both simulated and real-world experimental results demonstrate that, in complex surface environments characterized by severe decorrelation and low signal-to-noise ratios, the proposed estimator significantly outperforms existing methods. It not only effectively mitigates the estimation bias of the coherence matrix， but also substantially improves the accuracy of time-series consistent phase retrieval.

Conclusions The proposed method effectively reconstructs dense deformation phase fringes, increases the number of identifiable measurement points, and enhances the effectiveness and accuracy of subsequent phase unwrapping. As a result, it reduces monitoring blind spots and improves the detection capability of deformation in low-coherence environments.