Deciphering complex landslide kinematics through DInSAR wrapped phase stacking

Deep-seated landslides are widespread in mountain belts, and creep for long periods affecting large rock slopes and posing risks to human lives and infrastructures. They are controlled by rock type, structure, and progressive failure processes, and exhibit complex deformation patterns characterized by kinematic segmentation, heterogeneity, and nested sectors which might be prone to collapse. Additionally, displacement of shallow debris often obscure signs of deeper movements. Mitigating the risk associated with deep-seated landslides requires detecting and characterizing spatial and temporal movement patterns over wide areas. Satellite SAR interferometry (InSAR) generated from Sentinel-1 has proven to be valuable to this aim, however, with some limitations. High-quality interferograms enable effective wrapped phase fringe interpretation and unwrapped displacement maps, offering a more continuous picture of landslide kinematics. However, they are susceptible to noise or unwrapping errors, especially in heterogeneous and segmented landslides, reducing their accuracy. Multitemporal methods such as Persistent Scatterer Interferometry (PSI) provide accurate velocity estimates at specific points, but often fail capturing spatial segmentation or signals of processes occurring at different timescales.

To address these issues, we propose a stacking approach that leverages wrapped InSAR interferograms generated with the ESA SNAP software. The method involves selecting temporal baselines suitable to capture the processes of interest based on geological constraints, generating and manually choosing multiple interferograms covering overlapping time windows, and calculating median stacked phase values and residuals for each pixel. As we aim at analyzing slow, permanent deformation, we assume that our target signals in single interferograms never reach 1-fringe (2.8 cm for Sentinel-1). We also developed ad hoc descriptors to test pixel-wise the validity of such assumption. This approach, implemented in the MATLAB™ script AMSTACK, was validated with synthetic interferograms simulating different landslide rheology, segmentation, and noise. The method was then applied to slow-moving rock slope deformations in Valfurva (Central Alps, Italy), where glacial valley flanks up to 1500 m high are carved into phyllites and mica-schists of Austroalpine tectonic units. These slopes exhibit structurally complex gravitational deformations with sharp morpho-structural features and nested rockslides in various stages of maturity. Using Sentinel-1 images from snow-free periods between 2015 and 2023, we generated over 120 interferograms with a 1-year temporal baseline, without applying APS corrections. The application of our stacking approach to manually-selected wrapped interferograms allowed to: a) enhance signal-to-noise ratios, quantifying displacement patterns, rates, and segmentation for specific slope sectors without unwrapping errors; b) distinguish shallow from deep-seated movements in InSAR signals; and c) identify nested sectors susceptible to catastrophic collapse. Validation with field and multitemporal InSAR data confirmed the method’s reliability. This provided robust interpretations where the slow permanent deformation occurs, while residuals offered additional insights into areas with high phase gradients, nonlinear temporal trends, and shallow mass movements.