ORIGINAL RESEARCH article
Front. Remote Sens.
Sec. Atmospheric Remote Sensing
Volume 6 - 2025 | doi: 10.3389/frsen.2025.1591276
Neural Network-Based Snow Depth Retrieval from AMSR-2 Brightness Temperatures Using ICESat-2 Measurement as Ground Truth
Provisionally accepted
- 1Analytical Mechanics Associates, Hampton, Virginia, United States
- 2Langley Research Center, National Aeronautics and Space Administration, Hampton, Virginia, United States
Select one of your emails
You have multiple emails registered with Frontiers:
Notify me on publication
Please enter your email address:
If you already have an account, please login
You don't have a Frontiers account ? You can register here
Using diffusion theory and Monte Carlo lidar radiative transfer simulations, Hu et al. (2022) and Lu et al. (2022) derived snow depth from moments of the lidar backscattering pathlength distribution and applied these methods to Ice, Cloud, and land Elevation Satellite-2 (ICESat-2) satellite lidar measurements. In this study, a neural network (NN) algorithm was developed to estimate snow depth using multiple channels from the Advanced Microwave Scanning Radiometer 2 (AMSR-2), humidity vertical 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 using ICESat-2 snow depth measurements, matched by time and geolocation, during the winter months of 2018-2019 over Arctic sea ice. The trained NN was then applied to AMSR-2 clearsky wide-swath data for the winters of 2018-2019 and 2019-2020, generating daily snow depth estimates across Arctic sea ice. Validation against independent ICESat-2 snow depth data demonstrated strong agreement, with a near-zero bias and a root mean square error (RMSE) of 10 cm. Additional validation was conducted using three approaches: (a) instantaneous time and location matching, (b) daily geolocation matching, and (c) monthly Arctic-wide averaged comparisons. For instantaneous matching, the NN-based AMSR-2 estimates exhibited minimal bias and an RMSE of 9 cm. In the daily matching approach, AMSR-2 showed a slight underestimation with a bias of approximately 3 cm and an RMSE of 9 cm. Similarly, in the monthly Arctic-wide comparisons, AMSR-2 estimates remained slightly thinner than ICESat-2, with a bias of about 1 cm and an RMSE of 10 cm. Overall, all validation methods demonstrated strong agreement between AMSR-2 and ICESat-2, confirming the reliability of this neural network-based approach. This study establishes the foundation for a long-term, daily monitoring system of snowfall, snowstorms, and blizzards, providing extensive spatial coverage of snow depth estimates across Arctic sea ice.
Keywords: Snow depth, AMSR-2, ICESat-2, Microwave, lidar, Neural Network, Pathlength distribution, multiple scattering
Received: 10 Mar 2025; Accepted: 23 May 2025.
Copyright: © 2025 Sun-Mack, Hu, Lu, Chen and Omar. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Sunny Sun-Mack, Analytical Mechanics Associates, Hampton, 23666, Virginia, United States
Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.