Context and objectives
The accurate monitoring of snow water equivalent (SWE) is critical to assess global water resources. More than a billion people worldwide rely on snow for the supply of drinking water. Snowmelt furthermore supports agriculture, industry, and hydropower generation, but can also cause hazards such as floods and avalanches. Snow also impacts our climate by reflecting solar radiation and reducing ground heat exchange with the atmosphere. Despite the importance, we still lack a basic understanding of how much snow is seasonally stored on Earth. This is especially the case in mountainous regions, where in situ measurements are unable to capture the spatial variability, land surface model simulations suffer from poor precipitation inputs, and operational passive microwave satellite retrievals saturate in deep snow. Active microwave observations at high frequencies (e.g., Ku-band) are promising, but satellite observations are lacking. Recently, a novel approach for satellite retrieval of snow depth (closely related to SWE) was developed in the frame of the BELSPO C-SNOW project. More specifically, radar backscatter observations at C-band from the ESA and Copernicus Sentinel-1 (S1) mission were used in a change detection algorithm to map weekly snow depth across mountain ranges of the Northern Hemisphere, with demonstrated accuracy. However, the radar measurements are impacted by a large array of variables (e.g., topography; soil; vegetation; snow microstructure, layering and wetness) that are difficult to simultaneously address in a simple change detection retrieval approach. The central hypothesis of Snowtrane is that Machine Learning Algorithms (MLA) allow for improved S1 retrievals, especially when also auxiliary EO data sets are included as input to further explain the scattering processes at hand. The objectives of the project are: (1) to improve the S1 retrievals of mountain SWE beyond the state-of-the-art by developing novel machine learning algorithms (MLA), and (2) to identify the optimal MLA and input data sets to achieve improved S1 snow retrievals.
Expected scientific results
1) A novel methodological framework based on machine learning for the remote sensing of snow depth and SWE in mountainous regions.
2) Recommendations on the optimal machine learning algorithm configuration for snow retrieval, including the type of algorithm, the architecture, and optimal combination of input EO data sets.
Societal and environmental relevance
The enhanced satellite-based SWE observations from Snowtrane are expected to benefit a series of cross-disciplinary scientific fields, including hydrology, glaciology, studies on wildlife migration and water-energy-carbon cycle interactions and can serve major applications as flood forecasting, numerical weather prediction and hydropower generation. By developing a retrieval approach for SWE from S1 (an ESA flagship mission with assured long-term continuity), Snowtrane can help establishing the foundation for the future monitoring of climate change impacts on snow water resources.
Expected products and services
A novel state-of-the-art dataset of S1 snow depth and SWE retrievals, at 500 m spatial and sub-weekly temporal resolution over the entire Alps for the period 2015-2023.
Space agencies (ESA and NASA), weather forecasting centers (ECMWF and NOAA), research institutes (e.g., studying the relationship between climate and snow water resources, water-energy-carbon cycle interactions, wildlife migration, flood monitoring, numerical weather prediction), water managers, avalanche forecasting centers, the agricultural sector, industry, the hydropower sector.