Abstract
The Snow Cover Fraction (SCF)—defined as the percentage of a pixel's surface covered by snow—is a key metric for characterizing snow distribution, particularly in areas with patchy or discontinuous snow cover. Its relevance is heightened in complex mountainous terrains or when analyzing satellite imagery with coarse spatial resolutions. Unlike binary Snow Cover Area (SCA) classifications, SCF offers finer granularity, enabling detailed snow characterization critical for hydrological and climate studies. SCF retrieval is predominantly based on optical remote sensing in the visible and shortwave infrared regions. While traditional methods, such as regression on the Normalized Difference Snow Index (NDSI) and multispectral unmixing, are widely used for their simplicity and reasonable performance, they face significant limitations in regions with complex topography or under atmospheric disturbances. Furthermore, these approaches fail to leverage the full potential of hyperspectral data, which offers richer spectral information. This study investigates the performance of linear and non-linear spectral unmixing methods for SCF retrieval using hyperspectral PRISMA and multispectral Sentinel-2 imagery. Data were acquired over Cervinia, Italy, on 4 July 2024, during a dedicated in situ campaign. Field spectroscopy data and very high-resolution (VHR, 25 cm) RGB images from a drone were collected to generate reference SCF maps at 20 m and 30 m resolutions for validating PRISMA and Sentinel-2 estimations, respectively. SCF retrieval algorithms evaluated in this work include linear regression on NDSI, linear spectral unmixing, and non-linear spectral unmixing. Various combinations of end-member spectra, sourced both from in situ measurements and direct image extraction, were tested. Additionally, the hyperspectral capabilities of PRISMA enabled experimentation with diverse spectral band combinations and bandwidths. Preliminary results highlight that SCF below 20% is challenging to detect. Nonetheless, the rich spectral information provided by PRISMA demonstrates improved performance compared to multispectral sensors. However, geolocation accuracy significantly impacts the retrieval results, underscoring the need for precise alignment in image processing. This work represents one of the first real-world validations of SCF retrieval methods using both hyperspectral and multispectral sensors. The findings enhance our understanding of SCF detection limits and the sensitivity of retrieval algorithms, contributing to advancements in snow monitoring techniques. The insights gained may inform the design of future multispectral sensors optimized for SCF retrieval. Acknowledgements • “Research work carried out using ORIGINAL PRISMA Products - © Italian Space Agency (ASI); the Products have been delivered under an ASI License to Use” . • This work is carried out within Contract “SCIA” n. 2022-5-E.0 (CUP F53C22000400005), funded by ASI in the “PRISMA SCIENZA” program.