Improved snow depth retrieval over Arctic sea ice from FY-3 satellites using machine learning and its future scenario projection
(2024) In Journal of Hydrology 644.- Abstract
Snowpack atop sea ice plays a vital role in the Arctic heat and energy balance by reducing the energy transfer between the atmosphere and the ocean. This study presents an improved snow depth retrieval model over Arctic multiyear ice (MYI) from the Microwave Radiometer Imager (MWRI) data of the Fengyun-3 (FY-3) series satellites using the optimal machine learning approach identified. Validation against the Operation IceBridge (OIB) data revealed our snow depth estimates had a root mean square error (RMSE) of 7.88 cm and 6.83 cm for snow depth estimates over first-year ice (FYI) and MYI, respectively. FY-3 derived snow depth estimates from 2010 to 2014 were used to evaluate snow depth outputs from the Coupled Model Intercomparison... (More)
Snowpack atop sea ice plays a vital role in the Arctic heat and energy balance by reducing the energy transfer between the atmosphere and the ocean. This study presents an improved snow depth retrieval model over Arctic multiyear ice (MYI) from the Microwave Radiometer Imager (MWRI) data of the Fengyun-3 (FY-3) series satellites using the optimal machine learning approach identified. Validation against the Operation IceBridge (OIB) data revealed our snow depth estimates had a root mean square error (RMSE) of 7.88 cm and 6.83 cm for snow depth estimates over first-year ice (FYI) and MYI, respectively. FY-3 derived snow depth estimates from 2010 to 2014 were used to evaluate snow depth outputs from the Coupled Model Intercomparison Project Phase 6 (CMIP6). Based on evaluation results, seven CMIP6 models were selected for snow depth predictions, namely ACCESS-ESM1-5, AWI-CM-1-1-MR, CAMS-CSM1-0, CanESM5, FGOALS-f3-L, MRI-ESM2-0, and NorESM2-MM. Cold season snow depths from 2015 to 2100 were generated under four shared socioeconomic pathway-representative concentration pathway (SSP-RCP) scenarios (SSP126, SSP245, SSP370, and SSP585). The results indicated that short-term increases in radiation forcing have minimal impact on snow depth variability; however, as time goes by and radiative forcing accumulates, the rate of snow depth decrease accelerates. Relative to the average snow depth in the baseline period (2010–2014), the percentage change in snow depth during the late 21st century (2081–2100) under the SSP585 scenario (−63.53 %) is more than twice the percentage change observed under the SSP126 scenario (−29.63 %). These findings highlight the potential for substantial snow loss over Arctic sea ice under high greenhouse gas emission scenarios.
(Less)
- author
- Ke, Chang Qing ; Li, Haili ; Shen, Xiaoyi ; Wang, Zifei ; Shi, Lijian ; Ludwig, Ralf and Duan, Zheng LU
- organization
- publishing date
- 2024-11
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Arctic sea ice, CMIP6, FY-3 brightness temperature, Machine learning, Snow depth, Spatiotemporal variation
- in
- Journal of Hydrology
- volume
- 644
- article number
- 132105
- publisher
- Elsevier
- external identifiers
-
- scopus:85205935997
- ISSN
- 0022-1694
- DOI
- 10.1016/j.jhydrol.2024.132105
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: © 2024 Elsevier B.V.
- id
- 34a5571d-c91b-483b-8213-384e837e20f9
- date added to LUP
- 2024-11-27 11:17:24
- date last changed
- 2025-05-29 02:15:58
@article{34a5571d-c91b-483b-8213-384e837e20f9,
abstract = {{<p>Snowpack atop sea ice plays a vital role in the Arctic heat and energy balance by reducing the energy transfer between the atmosphere and the ocean. This study presents an improved snow depth retrieval model over Arctic multiyear ice (MYI) from the Microwave Radiometer Imager (MWRI) data of the Fengyun-3 (FY-3) series satellites using the optimal machine learning approach identified. Validation against the Operation IceBridge (OIB) data revealed our snow depth estimates had a root mean square error (RMSE) of 7.88 cm and 6.83 cm for snow depth estimates over first-year ice (FYI) and MYI, respectively. FY-3 derived snow depth estimates from 2010 to 2014 were used to evaluate snow depth outputs from the Coupled Model Intercomparison Project Phase 6 (CMIP6). Based on evaluation results, seven CMIP6 models were selected for snow depth predictions, namely ACCESS-ESM1-5, AWI-CM-1-1-MR, CAMS-CSM1-0, CanESM5, FGOALS-f3-L, MRI-ESM2-0, and NorESM2-MM. Cold season snow depths from 2015 to 2100 were generated under four shared socioeconomic pathway-representative concentration pathway (SSP-RCP) scenarios (SSP126, SSP245, SSP370, and SSP585). The results indicated that short-term increases in radiation forcing have minimal impact on snow depth variability; however, as time goes by and radiative forcing accumulates, the rate of snow depth decrease accelerates. Relative to the average snow depth in the baseline period (2010–2014), the percentage change in snow depth during the late 21st century (2081–2100) under the SSP585 scenario (−63.53 %) is more than twice the percentage change observed under the SSP126 scenario (−29.63 %). These findings highlight the potential for substantial snow loss over Arctic sea ice under high greenhouse gas emission scenarios.</p>}},
author = {{Ke, Chang Qing and Li, Haili and Shen, Xiaoyi and Wang, Zifei and Shi, Lijian and Ludwig, Ralf and Duan, Zheng}},
issn = {{0022-1694}},
keywords = {{Arctic sea ice; CMIP6; FY-3 brightness temperature; Machine learning; Snow depth; Spatiotemporal variation}},
language = {{eng}},
publisher = {{Elsevier}},
series = {{Journal of Hydrology}},
title = {{Improved snow depth retrieval over Arctic sea ice from FY-3 satellites using machine learning and its future scenario projection}},
url = {{http://dx.doi.org/10.1016/j.jhydrol.2024.132105}},
doi = {{10.1016/j.jhydrol.2024.132105}},
volume = {{644}},
year = {{2024}},
}