首页
>
期刊首页
>
海洋学报(英文版)
>
2026年7期
>
Reconstruction of Arctic sea ice snow depth during the freezing season from dual-frequency satellite altimetry using machine learning(2010-2024)
Reconstruction of Arctic sea ice snow depth during the freezing season from dual-frequency satellite altimetry using machine learning(2010-2024)
摘要
Snow on Arctic sea ice significantly influences climate processes and represents the main source of uncertainty in satellite altimetry-based sea ice thickness retrievals.In this study,we develop SnowModel-LGB,a monthly snow depth reconstruction model for Arctic sea ice during the freezing season(October to April),based on a Light Gradient Boosting Machine(LightGBM)framework.The model uses LaKu snow depth(retrieved from ICESat-2 and CryoSat-2 dual-frequency satellite altimetry)as the target variable,incorporating atmospheric reanalysis data,sea ice parameters,and spatiotemporal features as predictors.The model is then applied to reconstruct monthly snow depth over Arctic sea ice at 25 km resolution during freezing seasons from 2010 to 2024.Results show that independent evaluation against April 2019 Operation IceBridge(OIB)observations yields strong performance(R=0.72,RMSE=5.2 cm),consistent with the LaKu snow depth validation results(R=0.81,RMSE=5.2 cm).For the extended 2010-2018 period,SnowModel-LGB demonstrates strong agreement with OIB observations(R=0.64-0.72,RMSE=6.2-7.4 cm).Comparison with Snow and Ice Mass Balance Array(SIMBA)data from the MOSAiC expedition demonstrates consistent seasonal variability,with monthly mean comparisons yielding a mean difference of -0.5 cm and RMSE of 1.3 cm.The reconstructed snow depth exhibits strong agreement with LaKu observations in terms of spatial distribution patterns.From a temporal perspective,multiyear monthly averaged values demonstrate seasonal progression from 9.7 cm in October to 19.4 cm in April across all ice types,with multiyear ice(15.1-26.7 cm)consistently exceeding first-year ice(5.9-17.2 cm).Compared to the Warren Climatology,the snow depth over Arctic sea ice has significantly decreased but has remained relatively stable from 2010 to 2024,indicating a new equilibrium state.The SnowModel-LGB extends LaKu observations to periods without dual-frequency altimetry,generating monthly snow depth time series for enhanced variability analysis and future operational applications.
年,卷(期):
2026, 45(7)
机标分类号:
G43
R338.1
TP317.4
在线出版日期:
2026年10月20日