Using AI for mapping calving front positions

June 21, 2021

For extracting frontal positions, we teamed up with the “Artificial Intelligence for Cold Regions” (AI-CORE) project. Erik Loebel from the team at TU Dresden explains the efforts in implementing AI methods for monitoring northeast Greenland glaciers.

What is the scientific background?

An accurate parameterization of iceberg calving is essential for understanding glacier dynamics, separating underlying processes, and constraining ice-sheet models. The increasing availability and quality of remote sensing imagery enable us to realize a continuous and precise mapping of relevant parameters such as calving front locations. However, the huge amount of data also accentuates the necessity for automated and scalable analysis strategies.

How do you detect changes in the calving front position and what are the challenges?

We develop an automated workflow for extracting calving front positions from optical Landsat-8 imagery utilizing deep learning methods. The workflow is based on semantic image segmentation using a Convolutional Neural Network as well as a unique set of multi-spectral, textural and topographic input features. This enables us to provide exceedingly dense calving front datasets for most of the Greenlandic outlet glaciers like the Zachariæ Isstrøm and 79° North Glacier. Robust model predictions are especially challenging in this area since calving, ocean, as well as illumination conditions change drastically throughout the year.

What can we learn from the result?

The animation presents preliminary results of our work showcasing over 350 frontal positions for the 79° North Glacier from 2013 to 2021. Along with other geodetic measurements, the derived frontal positions will be part of the GROCE subproject 6 and support to separate various effects regarding ice dynamics and surface processes of the Greenland Ice Sheet.

More information about automated calving front delineation: