Production of a High-Precision Daily Cloud-Free Snow Cover Fraction Product at 5 km Resolution for the Northern Hemisphere (1981-2024)

Accurate snow cover information is crucial for studying global climate and hydrology. Existing snow cover fraction products struggle to balance temporal coverage and spatial resolution. We propose a new method to produce daily cloud-free SCF products at a 5 km resolution for the Northern Hemisphere from 1981 to 2024. This approach integrates advanced techniques such as asymptotic radiative transfer (ART), physics-constrained deep learning, stacked ensembles, and multi-level decision trees. Specifically, we develop a deep learning algorithm for SCF retrieval based on enhanced resolution passive microwave data (6.25 km), considering brightness temperature, soil properties, and land cover types. A cloud discrimination algorithm using a multi-level decision tree based on AVHRR data is constructed to improve the ability to distinguish between snow and clouds in medium-resolution optical remote sensing data. By utilizing surface reflectance remote sensing data, terrain data, and meteorological reanalysis, we establish a physics-constrained deep neural network model to accurately estimate SCF. Furthermore, we develop different fusion strategies for SCF in cloudy and cloud-free regions based on microwave and optical remote sensing, employing deep learning algorithms and ensemble learning techniques. This product is expected to better serve global climate, hydrological, and related research.