A Network-Based Phase-Gradient Stacking Method for Resolving Long-Wavelength Deformation From Low-Coherence SAR Interferograms

Abstract

Since the early 1990s, Synthetic Aperture Radar Interferometry (InSAR) has significantly advanced surface deformation measurement across various applications. Despite the successes, InSAR faces challenges in retrieving long-wavelength deformation, particularly in vegetated regions. This is primarily due to the tropospheric phase delays and unwrapping errors. Here we propose a network-based phase-gradient stacking (NPG-Stacking) method that calculates and stacks the phase gradients based on a triangular network connecting iteratively selected high-coherence, residue-free pixels. Afterward, we apply the weighted least squares inversion to retrieve the deformation phase from the stacked phase gradients, during which a posterior test is iteratively conducted to refine the network. Based on these procedures, the NPG-Stacking allows for reducing tropospheric delays without unwrapping individual interferograms. We validate the NPG-Stacking using C-band Sentinel-1 data in three InSAR-challenging scenarios associated with earthquake cycle deformation: the far-field postseismic deformation following the 2011 Tōhoku earthquake, the interseismic deformation along the Median Tectonic Line (MTL) of Japan, and the coseismic deformation of the 2018 Anchorage earthquake in Alaska, where conventional InSAR methods largely failed to produce useable deformation fields. In all the cases, the proposed NPG-Stacking method reveals deformation patterns consistent with the ones interpolated from dense GNSS measurements, with corresponding accuracies of 4.5 mm/yr, 1.35 mm/yr, and 4.93 mm, respectively. Although the NPG-Stacking methods may be limited to retrieving long-wavelength deformation with simple temporal behaviors, the results demonstrate its robustness in low-coherence regions, highlighting its potential to extend the InSAR applicability in challenging environments where conventional methods may fail.