Large-scale Automatic Calving Front Segmentation and Frontal Ablation Analysis of Arctic Glaciers using Synthetic-Aperture Radar Image Sequences

• Applicants: Dr.-Ing. Vincent Christlein; Dr. Thorsten Seehaus
• Subject Area: Physical Geography ; Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: since 2023
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 512625584

The arctic glaciers are strongly affected by global climate change, in particular due to the Arctic amplification phenomenon. Thus, they contribute significantly to current sea level rise. There exists only sparse information on glacier front variations as well as the evolution of frontal ablation and its drivers for the huge amount of tide water glaciers in the Arctic. However, the steadily increasing amount of available remote sensing data, in particular since the launch of the Sentinel satellites within the Copernicus Program by ESA and the European Commission, facilitates large-scale spatio-temporal monitoring of glacier changes. Within the proposed project, we aim at applying deep learning techniques to automatically map calving front positions of arctic tidewater glaciers, by means of Sentinel-1 SAR imagery. Therefore, we will implement and evaluate different deep learning segmentation models, including also time-series information from SAR image sequences. To improve the robustness and performance of our segmentation results, we will enlarge the field-of-view by incorporating attention mechanisms, apply multi-task learning, and fine-tune our best model in a semi-supervised manner on glaciers not yet included in the training database. Uncertainty bounds will increase the interpretability of our model’s results. For the best performing approach, a fully automated operational processing pipeline will be implemented and applied on the huge amount of available Sentinel-1 imagery throughout the Arctic. Subsequently, the obtained information on glacier front changes will be combined with information on ice flux at the termini and climatic mass balance data to estimate frontal ablation. By including information on total mass balances of the glaciers and climatic mass balance data from regional climate modeling or down-scaling of reanalysis data, the contribution of frontal and surface mass balances to the total mass balance will be estimated. Finally, the obtained results on glacier changes in combination with information on atmospheric, oceanic, sea ice, and glacier geometry parameters will be analyzed using multivariate statistical approaches to identify the drivers of the observed glacier changes. The project results will enhance our understanding of the ongoing processes and cross-links between the different spheres. Moreover, we will obtain fundamental reference data for numerical glacier modeling, subsequently leading to improved projections of glacier evolution and sea-level changes.

• DFG Programme: Research Grants