Please note that this session was withdrawn and is no longer available in the respective programme. This withdrawal might have been the result of a merge with another session.
CR2.5 | Machine Learning for Large-scale Observation of the Cryosphere and Associated Natural Hazards in Mountainous Regions
PICO
Machine Learning for Large-scale Observation of the Cryosphere and Associated Natural Hazards in Mountainous Regions
Convener: Rodrigo Caye DaudtECSECS | Co-conveners: Flora WeissgerberECSECS, Elisabeth D. Hafner-AeschbacherECSECS, Manu TomECSECS
The cryosphere is central to many research topics and applications such as hydrology, glaciology, avalanche research and snow and mountain ecology. Understanding the storage of water in mountain glaciers and snow is essential for managing hydrological resources in many regions. Therefore, studying the processes and changes to the cryosphere that are affected by a changing climate is crucial for many populations worldwide. Remote sensing and state-of-the-art data analysis techniques are powerful tools for scientific observation and analysis of the cryosphere.

In this session, we invite abstracts that focus on the application of machine learning for remote sensing of the cryosphere and associated natural hazards in mountainous regions. We especially encourage contributions that observe and/or monitor variables such as snow depth, lake ice, glacial lakes, glaciers (incl. rock glaciers), permafrost, snow avalanches, etc. Submissions that use a wide range of input data, such as optical imagery (Sentinel-2, MODIS, VIIRS, Landsat-8 etc.), SAR imagery (Sentinel-1, TerraSAR-X, RADARSAT-2 etc.) or radar altimetry (CryoSat-2, SARAL/AltiKa), are expected. We highly encourage contributions that incorporate large-scale analyses (both spatially and temporally, e.g., country-scale or multi-year analyses) and/or multi-sensor data fusion.

Submissions that deal with a plethora of learning strategies ranging from classical machine learning to deep learning are expected. We strongly encourage submissions which explore approaches that use minimal reference data, such as weakly-, self-, or semi-supervised learning strategies, taking advantage of the large amount of available satellite data archives. In addition, other approaches which aim to be deployed at large scale, such as domain adaptation, will be well received. Methodologies that employ fully-supervised learning, multi-task learning, physics-inspired machine learning etc. are also welcome. We expect contributions with a meaningful validation/ comparison to reference data and a discussion about the limitations of the chosen methodology.

The session also offers the possibility to present results from ongoing research projects.