When one envisions Antarctica, the mind conjures up images of vast, snow-covered plains. However, hidden within this frozen expanse lie the mesmerizing blue-ice areas (BIAs), with their striking azure hue.
Despite occupying a small fraction of the continent, these BIAs hold immense scientific value. They not only ensnare precious meteorites but also play a crucial role in accelerating the melting of the Antarctic ice sheet. Additionally, they harbor ancient ice that can unravel Earth’s climatic history.
Uncovering these BIAs across the colossal expanse of Antarctica may seem like an insurmountable challenge. However, a groundbreaking application of machine learning, spearheaded by glaciologist Veronica Tollenaar and her team at Université libre de Bruxelles, has triumphed in this daunting task.
Their recent paper showcases an innovative approach that surpasses previous methods, which often led to errors in identifying BIAs. By distinguishing between temporary and enduring snow cover, this cutting-edge model has revolutionized the discovery of these enigmatic blue ice realms.
The researchers used a convolutional neural network to process data through multiple layers of filters, simulating how the human brain perceives and interprets visual information. The neural network was provided with inaccurate maps and a smaller number of maps with errors removed by hand. By learning from the “clean” maps, the model was able to detect mistakes in the inaccurate ones.
“Because we are using deep learning, [we] are able to abstract the immediate data, take special context into account, and make sure that we learn something that is more than a simple optical difference,” said Benjamin Kellenberger, a postdoctoral researcher at Yale and a co-author of the study.
Kellenberger’s findings confirm the power of detailed analysis and machine learning in improving the accuracy of Antarctic maps. By examining areas where discrepancies existed between the original and new maps, the research revealed a significant reduction in cases of missed or incorrectly identified blue ice. This breakthrough underscores the potential of merging geophysical and computational sciences to gain profound insights into our planet’s history.
The monumental scale of Antarctica demands innovative approaches, and this study paves the way for future advancements in mapping Earth’s evolution.
Journal reference:
- Veronica Tollenaar, Harry Zekollari, Frank Pattyn, Marc Rußwurm, Benjamin Kellenberger, Stef Lhermitte, Maaike Izeboud, Devis Tuia. Where the White Continent Is Blue: Deep Learning Locates Bare Ice in Antarctica. Geophysical Research Letters, 2024; DOI: 10.1029/2023GL106285
