Deep learning framework improves snow cover fraction estimation

Accurate snow cover information is crucial for studying global climate and hydrology. However, deep learning approaches for retrieving snow cover fraction (SCF) often suffer from limitations in training data dependence and interpretability.

A research team led by Hao Xiaohua from the Northwest Institute of Eco-Environment and Resources (NIEER) of the Chinese Academy of Sciences, in collaboration with Lanzhou University and Lanzhou Jiaotong University, proposed a novel deep learning framework for accurate snow cover fraction estimation, named ART-DL SCF model.

The study, titled “Estimating AVHRR Snow Cover Fraction by Coupling Physical Constraints into a Deep Learning Framework,” was published in ISPRS Journal of Photogrammetry and Remote Sensing on September 12.

This model couples the asymptotic radiative transfer (ART) model with advanced very high-resolution radiometer (AVHRR) surface reflectance data to retrieve the Northern Hemisphere SCF.

The ART-DL SCF model showed high accuracy, robustness, and generalization in both temporal and spatial in tests, performing well even in forested areas.

The model also performed favorably compared with the available SnowCCI AVHRR products from the European Space Agency (ESA).

The inclusion of physical constraints not only enhanced the estimation accuracy and stability but also mitigates underestimation issues.

This study presents a new model for retrieving SCF that shows great potential in generating daily long-series SCF and snow cover extent (SCE) products with high accuracy in the Northern Hemisphere. It is expected to provide valuable input for various fields, including hydrology, ecology, and atmospheric science, according to the researchers.