SnowGalileo: Daily Gap-Free 100 m Fractional Snow Cover Mapping in Mountain Regions from Multimodal Earth Observation Data

Kurzfassung

Mountain snow is an important component of the cryosphere and directly affects downstream livelihoods. Accurate monitoring of mountain snowpacks remains challenging because of complex topography and frequent cloud cover. Combining data from multiple Earth observation (EO) satellites can improve spatial and temporal coverage, but extracting fractional snow cover (FSC) from sensors with different spatial and temporal resolutions remains difficult. AI-based Earth foundation models can process diverse sensor inputs and have been successfully applied to various remote sensing tasks; however, they often lack cryosphere-specific pre-training data. Here, we present SnowGalileo, the first model designed to integrate multimodal EO data with varying spatial and temporal resolutions, and evaluate its potential for generating daily, gap-free FSC maps at 100 m resolution, particularly in mountainous regions. SnowGalileo combines one week of satellite time-series data from multiple sensors on board of Sentinel-1, Sentinel-2, Sentinel-3, Landsat, Terra/Aqua and S-NPP/NOAA-20/21, with additional information such as topography, land cover, and meteorological data to predict FSC for a given day. The model is trained on 150,000 pre-training samples from mountain ranges across the Northern Hemisphere and evaluated in the Canadian Rockies and the Swiss Alps. SnowGalileo consistently outperforms standard machine learning approaches such as random forests and support vector regression. Results show robust performance under cloudy conditions, time-series gaps, and limited availability of high-resolution imagery in mountain regions. This approach could ultimately enable the generation of continuous, gap-free FSC time series for mountain regions worldwide.