Tracking landslide terrain motion with Very High Resolution optical image time series.

Times series of VHR optical imagery (SPOT6-7, SPOT-HRS Pléiades, PNEO), with their high spatial resolution (<0.5 to 2 m) and stereoscopic capabilities are offering huge potential for monitoring surface deformation using Optical Image Correlation (OIC) techniques. Very-High spatial resolution allows to enhance both the sensitivity and the accuracy of the measurements leading to the detection of small changes in deformation rates  (possibly close to 0.10 m in theory) for Pléiades imagery. However, the exploitation of these VHR satellite image time series remains challenging because of errors associated with the image acquisition geometry, which are potentially high in mountain regions with complex and string topography.

We propose an automated and generic processing chain, based on the GDM-OPT workflow (Provost et al., 2022) initially tailored for Sentinel-2 (10 m spatial resolution) image time series in order to process time series of VHR imagery, taking into account Pléiades Panchromatic monoscopic and stereoscopic data products. 

The approach consists first in the generation of intermediary DSM by a classical stereo-photogrammetric process. Second, in order to compensate for the planimetric and vertical errors, we correct the generated DSMs through an alignment to a reference topography. We then compute the ground coordinates of tie points of the image system taking into account the newly aligned topography. Considering these points as GCPs (Ground Control Points) and by performing a new bundle adjustment forced to fit to them, the alignment step is integrated in the stereo-photogrammetric process. Then, a new DSM and an ortho-image mosaïc consistent with the reference topography are calculated. Finally, the ortho-image mosaïcs are correlated using a specific pairing network (Stumpf et al., 2017). At the end of this step, all the displacement maps obtained (North-South, East-West) are inverted into a displacement time series. 

The processing workflow is tested on the two landslides of La Valette and Aiguilles/Pas de l’Ours (where time series of 8 Pléiades imagery are available) allowing to retrieve the mean velocity and the ground displacement time series for each pixel. We validate the proposed workflow by comparing the results of the processing chain and in-situ dataset (GNSS, LiDAR and photogrammetry). We show that the proposed methodology allows the monitoring of large landslides displacement, with velocity larger than 0.07 m/year.