The need for more accurate snow data has long been felt in Switzerland, a country where winter tourism and hydropower play pivotal roles. Traditional methods relied heavily on a network of about 400 meteorological stations, which left much of the terrain under-monitored. Konrad Schindler, Professor of Photogrammetry and Remote Sensing at ETH Zurich, emphasizes the limitations of current methods, stating, "While the best snow maps currently available for Switzerland have an effective resolution of around 250 by 250 metres, our maps allow the viewer to zoom in to 10 by 10 metres to read the snow depth."
The technology leverages data from Sentinel-2 satellites operated by the European Space Agency (ESA), which capture highly detailed images of Earth's surface. By analyzing optical and infrared images with a resolution of 10 by 10 metres per pixel, the AI system can identify snow presence and track changes over time. This method represents a considerable leap from the spatial limitations of ground-based measurements.
However, determining snow depth from satellite imagery is not straightforward. It requires the integration of additional data sources to account for varying terrain. To address this, the ETH team incorporated detailed terrain data from swisstopo, the Swiss Federal Office of Topography, into their AI model. This data, reflecting terrain variations like slope and orientation, is crucial for understanding how snow distribution is affected by geographical features.
The AI system underwent a rigorous training process, using snow maps provided by ExoLabs and real measurements for calibration. The process, known as supervised learning, involved continuously refining the AI's estimates against actual data, enhancing its accuracy. The system was further fine-tuned with detailed snow data from the Dischma valley, collected by the Swiss Federal Institute for Forest, Snow and Landscape Research WSL. This fine-tuning allowed the AI to understand the micro-variations in snow depth caused by minor terrain differences.
This new standard for measuring snow depth was successfully tested over two winters, with promising results. Schindler confidently asserts, "We expect that this will set a new standard for measuring snow depths in Switzerland." The technology not only provides more precise and granular data but also includes a measure of certainty, adjusting its reliability based on factors like recent weather conditions and image availability.
ExoLabs, responsible for the technology's commercialization, has already integrated these high-resolution snow maps into various applications. These apps, including Outdooractive, Strava, Skitourenguru, Huttenbuch, and swisstopo, cater to a diverse user base, ranging from outdoor enthusiasts to professionals in weather forecasting and hydropower management.
Looking ahead, ExoLabs CEO Reik Leiterer envisions broader applications for this technology. Plans are underway to expand these enhanced snow maps to regions beyond the Alps, including Scandinavia, the Pyrenees, and parts of North and South America.
This collaboration between ETH Zurich and ExoLabs represents a creative convergence of satellite imagery, AI, and terrain data, providing a different, more accurate approach to snow depth measurement. This development not only benefits Switzerland but also sets the stage for global applications in snow monitoring, a crucial aspect of understanding and managing our natural environment.
Research Report:Snow depth estimation at country-scale with high spatial and temporal resolution
Related Links
ETH Zurich
It's A White Out at TerraDaily.com
| Subscribe Free To Our Daily Newsletters |
| Subscribe Free To Our Daily Newsletters |
