An Improved Antarctic Sea Ice Thickness Dataset Derived from CryoSat-2 Using LightGBM

Abstract

Abstract Sea ice is crucial for modulating Antarctic air–sea fluxes, and its thickness (SIT) is the primary factor controlling the exchange of heat, moisture, and momentum. Although CryoSat-2 is commonly used for SIT retrieval, conventional algorithms rely on empirical parameters and auxiliary data that introduce substantial uncertainties. In this study, we developed a novel SIT dataset for 2010–24, derived directly from radar parameters using the Light Gradient Boosting Machine (LightGBM) machine learning method. Intercomparisons show that the LightGBM-derived SIT shows better consistency with the ICESat-2 product than conventional algorithm results. Validation against shipborne observations indicates that LightGBM-based monthly gridded SIT achieves a mean absolute error of 0.558 m, which is lower than with conventional methods (0.823 m). Temporal comparisons reveal that the LightGBM-derived sea ice volume (SIV) exhibits a more realistic seasonal cycle, with the maximum value occurring in September, compared to the conventional method, which shows a peak in August. This new SIT dataset provides a robust basis for estimating SIV with reduced uncertainty, investigating sea ice variability mechanisms, and assessing the impact of sea ice changes.