Snow in the Alps (image: David Gubler via Flickr - CC BY-NC-SA)
Author profile picture

Swiss researchers at ETH Zurich have teamed up with ExoLabs to transform snow depth measurement across Switzerland. Their cutting-edge system zooms into a granular 10×10 metre resolution, a significant leap from the current 250×250 metre standard. The system leverages Sentinel-2 satellite data, refined by deep learning techniques. In the realm of snow measurement, they have introduced a level of detail previously unattainable. This is particularly impactful in Switzerland, a country where snow depth is not just a seasonal statistic but a critical factor for various sectors such as tourism and hydropower.

why you need to know this

Switzerland leverages satellite images, historical data, and machine learning to create a high-resolution snow depth map of the country. Another example in which machine learning and historical data enhance current data.

The technology behind the precision

The AI system employed in this innovation was meticulously trained using optical and infrared images from the European Space Agency’s Sentinel-2 satellites. These were supplemented with detailed terrain data from swisstopo, the Swiss Federal Office of Topography. The AI was then fine-tuned with highly detailed snow data collected from the Dischma valley in eastern Switzerland. This approach allowed for accurate snow depth estimates to be extrapolated across the entire country.

The system’s accuracy has been validated against high-fidelity snow depth maps obtained through airborne photogrammetry over a three-year period. This rigorous evaluation ensures that the maps produced are not only high-resolution but also reliable. The integration of probabilistic deep learning also means that uncertainty estimates are well-calibrated for each snow depth estimate, an important feature for applications where precision is critical.

What sets this new method apart is its ability to update snow depth maps regularly without the need for new ground measurements in fair weather conditions. This is a game-changer in terms of efficiency and resource management. The research team, with extensive experience in satellite image analysis, has created a system that generates higher-resolution snow maps for Switzerland than has ever been possible before. The artificial intelligence component is key to this advancement. It uses recurrent convolutional neural networks, a type of deep learning technology, to estimate snow depth with a high spatial resolution of 10 metres ground sampling distance (GSD), at a weekly frequency and large scale.

Impact on tourism and beyond

The implications of this development extend far beyond the academic. ExoLabs plans to commercialise the technology, offering high-resolution snow maps through various applications. This will aid not only in tourism but also in sectors such as insurance, hydropower generation, and risk management for winter sports. The accurate and timely information provided by the maps can help in planning and developing strategies to adapt to climate change.

Switzerland’s tourism industry, heavily reliant on winter sports, stands to benefit immensely. The availability of precise snow depth information will support destination marketing organisations (DMOs) in their promotional efforts and strategic planning. Given the potential impact of climate change on snowfall patterns, the ability to precisely monitor and predict snow cover is becoming increasingly important. This technology equips DMOs with the data necessary to adapt and potentially mitigate economic impacts due to fluctuating winter tourism.

Collaboration and funding

This project, dubbed “Deepsnow”, represents a successful public-private partnership. It was funded by Innosuisse, the Swiss Innovation Agency, and involved collaboration with the Swiss Federal Institute for Forest, Snow and Landscape Research (WSL), Outdooractive, and MountaiNow, among others. Such collaborations embody Innosuisse’s principal objective: fostering the symbiotic relationship between research and industry to bring innovative products to market.

The support from Innosuisse was not limited to financial backing. The project also benefited from active assistance during the proposal writing phase, demonstrating Innosuisse’s commitment to nurturing innovation from inception to market readiness.

As Reik Leiterer, CEO of ExoLabs, envisions, this technology is not confined to the Swiss Alps. There are plans to expand the service to other mountainous regions across the globe, such as Scandinavia, the Pyrenees, and the Americas. This foresight underlines the global potential of the Deepsnow project, indicating its scalability and adaptability to different geographical contexts.