Semi-automated detection of terrain stability in the Swiss Alpine periglacial environment from DInSAR scenes
Barboux, Chloé1; Delaloye, Reynald1; Lambiel, Christophe2; Strozzi, Tazio3; Raetzo, Hugo4; Collet, Claude1
1University of Fribourg, Geosciences departement, Geography Unit, SWITZERLAND; 2University of Lausanne, Institute of geography, SWITZERLAND; 3Gamma Remote Sensing, SWITZERLAND; 4Federal Office for the Environment FOEN, SWITZERLAND

A large spectrum of mass wasting processes (e.g. landslides, rock glaciers, debris-covered glaciers) are actively changing the surface topography of alpine mountain slopes over time. The rate of motion is typically ranging from millimeter to several meters per year. The significant quantities of debris delivered downstream mass wasting slope may drastically change the debris flow activities in the subjacent gully and modify locally the exposure of settlements and transportation systems to damages on alluvial fans in the valley bottom. As a consequence, mass wasting processes may pose a significant threat to infrastructures as well as for lives of people living and moving in alpine terrain. Surveying systems are today mainly based on field observations, historically recorded events as well as experience and expert knowledge. Several sites are surveyed by geophysical investigations, some of them are actually surveyed by annual or seasonal DGPS (Differential Global Positioning System) campaigns, and even more by webcams and GPS installed on-site.
Regionally, the detection and characterization of slope motion above the tree line in the Swiss Alps has been performed using a large set of DInSAR (differential synthetic aperture radar interferometry) scenes from 1991 to 1999 (Delaloye 2007a, 2007b, 2010). A large inventory of moving areas (hereafter called InSAR polygons) has been established in the periglacial belt. The exploitation of DInSAR data and the set up of inventory have been carried out by visual interpretation on the basis of a large set of interferograms comprising various time lapses (from 1 day to 3 years) and polygons outline areas where DInSAR signals corresponding to a possible slope movement have been detected (Delaloye 2007b, Lambiel 2008). Current inventory is divided into signals of different magnitude order (cm/day, dm/month, cm/month, cm/year) and contains signal patterns that are related to different mass wasting processes. The main objective now is to - so far as possible - automatically update inventory by integrating the most recent data to detect potential change in activity rate of landforms

The process presented here uses segmentation and classification methodologies applied on interferometric coherence and phase images to detect the stability of the terrain within DInSAR scenes. Then, the resulting stability map is used to update inventory of InSAR polygons. The method consists firstly in detecting noisy areas in the interferogram by thresholding the interferometric coherence. Then, a standard region growing algorithm is applied on the interferometric phase to segment the reminding areas into stable or moving areas. Several user-defined seed points are used to start the region growing process. The resulting map is thus classified into 3 classes: stable areas, coherent moving areas and noisy areas (due to motion or not). By applying this process to several interferograms having the same time lapse and by combining the classification results, we can improve the quality of the stability map showing the behavior of the terrain during the period described by the selected pairs. Pixels are thus classified into each of the 3 categories the most represented. An extra class is added indicating when the algorithm is not able to clearly classify the pixel. An additional map has been produced to give an index of reliability concerning the classification of each pixel point. This one is performed on the idea that the more the number of images used for the classification and the number of images classifying the pixel in one specific class are high, the more the classification is reliable.
The second step consists in detecting the potential change in the activity rate of InSAR polygons. Each of them is automatically compared with the stability map to see the proportion of its surface in each class. The change in the activity rate is detected thanks to several thresholds determined by the time lapse and the wavelength relative to the stability map.

The method was tested in a small studied area of 37 km2 using 7 pairs of TSX data scenes from summers 2010-2011 with 11 days time lapse and validated by comparing visually the result. False change detection is mainly due to external factors as vegetation, snow or atmosphere (where the signal is noisy), due to border effect in layover and shadow areas, as well as due to a change of the outline of the landform and are most cases related to a low reliability index of classification. We found the method works reliably with numerous user defined seed points (3 in the studied region). Furthermore, the model is simple but robust and the classification of the terrain is much more reliable by using several DInSAR pairs. However, comparing InSAR polygons to stability maps with larger time-lapse could improve the change detection especially for slow moving polygons. Future research is directed to widespread testing of the algorithm on data generated by various sensors and improvement of the algorithm in high noise regions. Moreover we would like to explore the ability to automatically identify and survey mass movement in other contexts by using the resulting stability map not only to (i) update inventories, but also (ii) to detect new moving areas, (iii) as well as to assist in the development of accurate inventory as a useful tool for visual interpretation of DInSAR data.

Acknowledgements
TERRASAR-X Data from LAN0411 and LAN1145 © DLR; DHM25 © 2003 Swisstopo, Swissimages, Pixel Maps 25 © 2010 Swisstopo (5701137809/000410, /000010).

Bibliography
Delaloye R., Lambiel C., Lugon R., Raetzo H., Strozzi T. (2007a). ERS InSAR for detecting slope movement in a periglacial mountain environment (western Valais Alps, Switzerland). High Mountain Proceedings of the 9th International Symposium on High Mountain Remote Sensing Cartography (HMRSC-IX), Graz, Austria, 14-15 Sept. 2006
Delaloye R., Lambiel C., Lugon R., Raetzo H., Strozzi T. (2007b). Typical ERS InSAR signature of slope movements in a periglacial mountain environment (Swiss Alps). Proceedings of the Envisat Symposium, Montreux, Switzerland, 23-27 April 2007.
Delaloye R., Strozzi T., Lambiel C., Barboux C., Mari S., Stocker A., Techel F., Raetzo H. (2010). The contribution of InSAR Data to the early detection of potentially hazardous active rock glaciers in mountain areas. Proceedings ESA Living Planet Symposium 2010, Bergen, Norway (ESA SP-686)
Lambiel C., Delaloye R., Strozzi T., Lugon R., Raetzo H. (2008). ERS InSAR for Assessing Rock Glacier Activity. Proc. of the 9th International Conference on Permafrost, Fairbanks, Alaska, USA, June 29 - July 3, 2008