TerraByte-DNN2Sim
On the trail of the mystery of the laws of calving
Ice mechanics is the core business of Dr. Angelika Humbert, Professor of Ice Modelling and Glaciology at the Alfred Wegener Institute (AWI) – the Helmholtz Centre for Polar and Marine Research in Bremerhaven. Her field of research: glaciers in Greenland and Antarctica.
“I have been biting my teeth for about a decade to develop a calving law,” she says. But despite intensive research, good simulations and highly sophisticated theories, Humbert and her colleagues have yet to predict where an iceberg will calve. “Glacier fronts are tricky,” says Humbert. “Because there are so many cracks in the ice at their edges that it’s hard to tell when and where it will break.” As an ice modeller, however, she needs such a law. After all, she wants to make predictions about changes in the glacier masses, not least to be able to forecast sea-level rise more accurately.
Satellite images and so-called deep neural networks (in other words: artificial intelligence in the form of neural networks) could be a key to this puzzle: In a joint project of the AWI and the German Aerospace Center (DLR), scientists are working on developing a model that succeeds in continuously tracking the glacier fronts in Antarctica and registering even the smallest movements using novel imaging algorithms, a complex data processing pipeline, mathematical optimisation and high-performance computers. Key of the anticipated achievement is that the analysis of the data and the simulation of the processes are conducted on the same compute infrastructure.
After all, somewhere in this diverse geometric information should be the answers that bring the researchers closer to discovering a law of calving. “And even if we don’t find it yet,” says Humbert, “this form of data analysis will get us to a much better baseline from which we can start calculating.”
The model they are developing could furthermore in the future be applied to other fronts whose behaviour is also of great importance to unravel: for example to algae or oil slicks on the ocean.
Publications
Other projects
HIT Permafrost
The Hidden Image of Thawing Permafrost
The project aims to develop a method for determining just how extensively thaw processes have already progressed in permafrost regions. The machine learning approach to be developed will be used to analyse radar images from aircraft in order to learn more about the properties of the subsurface permafrost.
SIM
Solar Image-based Modelling
The aim of the project is to develop an algorithm by which computers can automatically predict the space weather. This will make use of datasets of solar images that have been captured from space. The method could replace computationally demanding physics-based models and deliver space weather forecasts long before the effects of solar events are […]
SyNaToSe
Leveraging Cross-Domain Synergies for Efficient Machine Learning of Nanoscale Tomogram Segmentation
The aim is to develop an adaptable algorithm that can be used to perform different tasks in data and image analysis without needing to be trained with new, laboriously annotated images for each separate task.