Publications

Jacobs, L.,  Kervyn, M.,  Nobile, A.,  Poesen, J.,  Coleta, AJ. & Dewitte, O. 2016. ‘Synthetic stereo-mate generation for geomorphologic interpretation to improve landslide-susceptibility mapping in remote settings: Rwenzori Mountains national park’. Young Researchers' Overseas Day - Royal Academy for Overseas Sciences. Book of abstracts.

Conference abstract

In remote regions, reliable or detailed ancillary data such as lithology or fault structures, needed for landslide susceptibility assessment is often not available. To collect this data, geomorphological interpretation using stereo-viewing of aerial photographs or stereo satellite images is a powerful and often applied tool. However, due to a bi-modal rainy season and the general humid climate in equatorial Africa, the acquisition of optical remote sensing images in the stereoscopic mode is seriously hampered. A region where this lack of ancillary data and stereoscopic satellite images is particularly apparent is the Rwenzori Mountains on the border of Uganda and DR Congo. In this study we investigate the potential of synthetic stereo-viewing for geomorphological mapping and evaluate its added value for landslide susceptibility analysis. A synthetic ortho-mate pair was created by the Stereo Analyst extension in ArcGIS using a 5m TanDEM-X DEM (digital elevation model) as a topographic basis. By re-sampling the SPOT-6 NIR band, a synthetic stereo-mate image was created. The resulting stereoscopic information allowed the identification of faults, moraine deposits and glacial cirques for the entire Rwenzori Mountains region. In total 275 faults, 13 glacial cirques and 59 moraine deposits could be mapped. The smallest moraine deposit mapped measured 50 ha. In comparison, the most detailed lithological map previously available only contained 33 moraine deposits, the smallest measuring just below 80 ha. Furthermore, detailed fault maps or information on glacial cirques were previously unavailable. Finally, very large deep-seated landslides, which were not detected using monoscopic information but which leave clear topographic signatures became apparent through the stereoscopic interpretation. The collected thematic information was introduced into a pixel-based susceptibility model using logistic regression. The landslide inventory used here to build the susceptibility model is constructed using visual interpretation of 48 freely available very high resolution Google Earth images and one SPOT-6 image of 1.5m resolution covering the entire Rwenzori region. In total, it contains >500 landslides. Early results show that distance to faults, and the presence of glacial cirques and moraine deposits are all significant. While the generation of synthetic stereo-mates using monoscopic optical images and a detailed DEM is rarely used, this application could be applied in similar areas where the stereoscopic images are unavailable due to high cloud cover or where cost restraints limit their acquisition. As such this technique can help to extend landslide inventories and increase our understanding of the region's geomorphology.