AUTHOR=Sun-Mack Sunny , Hu Yongxiang , Lu Xiaomei , Chen Yan , Omar Ali 

TITLE=Neural network-based snow depth retrieval from AMSR-2 brightness temperatures using ICESat-2 measurement as ground truth

JOURNAL=Frontiers in Remote Sensing

VOLUME=Volume 6 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/remote-sensing/articles/10.3389/frsen.2025.1591276

DOI=10.3389/frsen.2025.1591276

ISSN=2673-6187

ABSTRACT=IndroductionEstimating snow depth over Arctic sea ice is essential for understanding climate processes and supporting operational forecasting. Previous work has demonstrated the use of lidar backscattering pathlength moments from Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) for snow depth retrieval. However, passive microwave sensors like the Advanced Microwave Scanning Radiometer 2 (AMSR-2) offer the potential for more frequent and spatially extensive observations.MethodsWe developed a neural network (NN) algorithm to estimate snow depth over Arctic sea ice using multi-channel brightness temperatures from AMSR-2, combined with humidity profiles and surface temperatures from the Global Modeling and Assimilation Office (GMAO) Goddard Earth Observing System for Instrument Teams (GEOS-IT) product. The NN was trained with temporally and spatially matched ICESat-2 snow depth data from the 2018–2019 winter season. The trained NN was then applied to AMSR-2 clear-sky wide-swath observations for the 2018–2019 and 2019–2020 Arctic winters, generating daily snow depth estimates across Arctic sea ice.ResultsValidation against independent ICESat-2 data showed strong performance: the NN-based AMSR-2 snow depth retrievals had a near-zero bias and a root mean square error (RMSE) of 10 cm. Further validation using (a) instantaneous matchups, (b) daily geolocation comparisons, and (c) monthly Arctic-wide averages confirmed consistent results. Instantaneous comparisons yielded a 9 cm RMSE with minimal bias, daily comparisons showed a 3 cm underestimation and 9 cm RMSE, and monthly averages exhibited a 1 cm bias and 10 cm RMSE.DiscussionThese results confirm the reliability of the neural network-based method for snow depth retrieval from AMSR-2. The approach enables daily, long-term monitoring of snow depth over Arctic sea ice, offering significant benefits for climate research and operational applications such as snowstorm and blizzard monitoring.