Enhanced understanding of Alpine mass movements Gathered through machine LEarning (EAGLE)

Automating mass movement identification with deep learning to improve data exploitation, to enable enhanced regional landslide insights and strengthening early warning systems.

We aim at harnessing recent deep learning advancements to automate the extraction of both location and activity status of mass movements from freely available, spaceborne remote sensing data. This allows to reduce human involvement and enhancing consistency in landslide identification, classification and monitoring. By combining interferometric data with optical imagery, we develop a scalable approach for continuous monitoring of slope instabilities. By automating mass movement identification with deep learning to enhance data exploitation, we strive towards advancing regional landslide insights and strengthening early warning systems.

eagle-image
Fig. 1: EAGLE Project vision.  A) Illustration of exemplary mass movement identification process as envisioned to be performed by deep learning network. Process classification following the definition by Hungr et al., 2014. Optical image: Google Hybrid. B) Manually mapped mass movements of a spatially filtered 12 days ifg. Colours separate countable fringe patterns and phase aliasing. Multi-looked average amplitude image across all amplitude images obtained during the processed time range. Uncertainty corresponds to the fringe count and movement detection uncertainty as perceived by the expert mapper. The image is displayed in radar projection 11.65 x 13.94 pixels in range and azimuth, respectively. C) A 12 days interferogram multiplied by corresponding signal coherence featuring a representation of the spatial scale we aim at. Background map: Bing Areal. Figure contains modified Copernicus Sentinel-1 data.

Contact

Gwendolyn Dasser
Lecturer at the Department of Earth and Planetary Sciences
  • NO G 69.3
  • +41 44 632 30 36

Professur für Ingenieurgeologie
Sonneggstrasse 5
8092 Zürich
Switzerland

JavaScript has been disabled in your browser