EGU2020-13782
https://doi.org/10.5194/egusphere-egu2020-13782
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Automatic glacier outlines extraction from Sentinel-1 and Sentinel-2 time series

Riccardo Barella, Mattia Callegari, Carlo Marin, Claudia Notarnicola, Marc Zebisch, Rudolf Sailer, Christoph Klug, Shtephan Galos, Roberto Dinale, and Stefano Benetton
Riccardo Barella et al.
  • Eurac Reserch, Earth Observation, Italy (riccardo.barella@eurac.edu)

Glaciers represent an important part of the hydrologic cycle in the Alps and they are very sensitive to climate change. Satellite remote sensing is an efficient tool for glacier monitoring because it provides a synoptic view over large areas. In literature, well-established methods for glacier delineation based on the Red and Short Wave Infrared (SWIR) ratio have been presented. These methods depend on a manual selection for each glacier of the “best scene”, i.e. absence of cloud coverage and minimum snow cover. A further manual refinement step is needed to handle possible errors, mainly due to cloud cover or shadows, and to include debris covered ice.

A manual approach for glacier outline extraction, especially if applied over large areas and beside the respective extraordinary amount of work, may be inadequate for at least two reasons:

1) The increased amount of available satellite data provided by the recently launched Sentinel-2 mission, which ensure at least one acquisition every 5 days on a given area;

2) The need for a more frequent update of the glacier outlines i.e. few years, due to the faster changes affecting glaciers during the last years.

In this work, we present an automatic method for glacier mapping, including bare ice and debris covered ice through the synergetic use of Sentinel-1 and Sentinel-2. The information of the Sentinel-2 time series is first classified with a Support Vector Machine (SVM) to detect cloud and snow. The snow and cloud masks are then used to select the non-cloudy pixels with the lowest snow coverage in the surrounding area. This is done by applying a moving window on the entire multi-temporal classified stack. The selected pixels for each band compose a multi-temporal cloud free mosaic, which represents the glaciers with the minimum snow cover for the current ablation season i.e., the “best scene”. If we compose the mosaic with classified pixels instead of the reflectance, we obtain the glacier – non glacier map that we use for outlines extraction. On the other hand, the Sentinel-1 coherence is used to detect the debris-covered ice over the areas classified as non-glacier from Sentinel-2. In detail, the Sentinel-1 time series is exploited to generate a multi-temporal coherence mosaic, which is representative of the loss of coherence due to the movement of the debris only. By properly thresholding this mosaic and considering the topographic information, the outlines of debris covered glaciers can be extracted.

The results obtained with the proposed method are compared with the recent official glacier inventory of South Tyrol (Italy) and Tyrol (Austria), which was derived from the manual interpretation of aerial orthophotos and lidar data by glacier experts.

How to cite: Barella, R., Callegari, M., Marin, C., Notarnicola, C., Zebisch, M., Sailer, R., Klug, C., Galos, S., Dinale, R., and Benetton, S.: Automatic glacier outlines extraction from Sentinel-1 and Sentinel-2 time series, EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-13782, https://doi.org/10.5194/egusphere-egu2020-13782, 2020

Display materials

Display file

Comments on the display material

AC: Author Comment | CC: Community Comment | Report abuse

Display material version 1 – uploaded on 30 Apr 2020, no comments