Project summary
30-month postdoc project
KU Leuven
funded by the Belspo Stereo program
Project description
Antarctica plays a critical role in regulating our planet’s climate and sea levels. One of the biggest scientific challenges today is accurately understanding how the Antarctic ice sheet is responding to a warming world—especially when it comes to processes like surface melting and snow accumulation. These processes, known as surface melt and surface mass balance (SMB), are key to predicting how fast ice will be lost and how much sea levels may rise. However, most climate models used today are not detailed enough to capture these changes accurately, especially in such a remote and complex environment.
The ClimaVision project aims to change that by developing advanced tools that use satellite data and artificial intelligence to create much sharper, more detailed maps of Antarctica’s climate. By combining cutting-edge deep learning methods with real satellite observations—such as images from Sentinel and MODIS missions—ClimaVision will be able to “zoom in” on climate data and uncover local changes that are currently missed by traditional models.
What makes ClimaVision unique is its use of physically informed machine learning. This means the AI models are not only trained to improve resolution, but also follow known scientific rules—like how temperature changes with elevation or how ice behaves under different conditions. This ensures the results are not just visually sharper, but scientifically reliable.
The improved climate data produced by ClimaVision will help scientists, policymakers, and climate agencies better understand how Antarctica is changing. It will also support better planning for the future—especially as the world works to prepare for rising seas and other impacts of climate change. All tools and results from the project will be made openly available, helping to drive further innovation in climate science and environmental monitoring.