Evaluating topography-based methods in 3D decomposition of InSAR 1D velocities obtained for translational landslides: Thompson River valley in Canada

Interferometric synthetic aperture radar (InSAR) has gained considerable attention as a landslide monitoring strategy owing to its high accuracy, large coverage, and relatively low associated costs. A crucial drawback of InSAR, however, has limited its further incorporation: one-dimensional estimations along the sensor's line-of-sight (LOS). This leads to an ambiguity in results and a less intuitive understanding of landslide kinematics. A frequently exercised approach to address this issue has been taking inspiration from the topography to establish compatibility assumptions between velocity components, yet little insight exists on the performance of these methods. The objective of this paper is to investigate the performance of four renowned topography-based methods-Surface Parallel Flow Model (SPFM), SPFM coupled with least-squares method (SPF-LSM), Aspect Parallel Flow Model (APFM), and Steepest Terrain Following Model (STFM)-in evaluating the magnitude and geometry of total velocity vectors. To this end, the analysis is performed on 202 Radarsat-2 and 243 Sentinel-1 scenes acquired over a section of the Thompson River valley, a critical railway corridor in Western Canada traversing 14 landslides. The results indicate that the APFM provides estimations with the lowest magnitude error (15-19 mm/yr or 18.75-23.75% of in situ measurements) compared to the other approaches. SPFM and SPF-LSM are highly sensitive to LOS variance and tend to bias the interpreted vectors toward the north orientation. However, APFM and STFM reflect more realistic aspect angles, with the former inclined to steeper travel angles and the latter suffering from erratic upward travel angles due to local topographies.