Unmanned Aerial Systems and DSM matching for rock glacier monitoring

Abstract

Among other techniques, aerial and terrestrial photogrammetry have long been used to control the displacements of landslides and glaciers as well as for the detection of terrain morphological changes. Unmanned Aerial Systems (UAS) are today an efficient tool to perform data acquisition in rough or difficult terrain, both safely and quickly, avoiding hazards and risks for the operators while at the same time containing the survey costs. Since 2012 ARPAVdA (the Regional Environmental Protection Agency of Aosta Valley, Italy) periodically surveys with UAS photogrammetry the Gran Sometta rock glacier, the Agency main monitoring site for the climate change impacts on high-mountain areas and related infrastructures.A Digital Surface Model (DSM) and an orthophoto of the rock glacier are produced after each survey flight. In order to accurately georeference them in a stable reference system, a Global Navigation Satellite System (GNSS) campaign is carried out at each epoch, to update the coordinates of signalised Ground Control Points (GCPs), since they partly lay in unstable (moving) areas. In late August 2015 a survey flight has been executed with a senseFly eBee RTK, with differential corrections sent from a ground reference station. The block has been adjusted without GCP using, as control information, only the projection centres coordinates encoded in the images. The RMS of the differences found on twelve Check Points were about 4cm in horizontal and 7cm in elevation, i.e. practically the same accuracy found using GCP. Differences between the DSMs produced at the same epoch with block orientation performed with GCP and with GNSS-determined projection centres were also investigated.To evaluate the rock glacier displacement fields between two epochs, corresponding features were at first manually identified on the orthophotos by a trained operator. To avoid the manual time-consuming procedure and increase the density of displacement information, two automatic procedures, the former using Least Squares Matching (LSM) and the latter a proprietary implementation of Semi-Global Matching (SGM) have been implemented. Both techniques were applied to pairs of orthophotos as well as to pairs of DSMs at different epochs. A discussion of the characteristics of the implemented methods is provided and the results of the comparison of the two methods with manual measurements are illustrated. Overall, results using DSM matching provided higher completeness of the displacement field than orthophoto matching, especially if long-term (year-to-year) comparisons are considered. At the same time, SGM in both cases produced less mismatches and more smooth and reliable displacement fields than LSM.