Estimation of Sea Ice Thickness Using FY-3E Data Based on Random Forest Method

Sea ice is a fundamental component of the cryosphere, necessitating a comprehensive understanding of its parameters. Notably, sea ice thickness (SIT) poses a significant challenge for accurate measurement. In this study, we employ a machine learning approach, specifically the Random Forest (RF) method, to estimate SIT by combining data from the Fengyun-3E (FY-3E) satellite and Soil Moisture Ocean Salinity (SMOS) datasets. This method relies on four input parameters: the incidence angle (θ), reflectivity (Γ), sea ice salinity (S), and sea ice temperature (T). The former two parameters are derived from FY-3E data, while the latter two are extracted from SMOS data. Moreover, FY-3E has the capability to receive signals reflected from both the Global Positioning System (GPS) and the Beidou Navigation Satellite System (BDS). Our investigation is centered on the Arctic polar region. An evaluation spanning from October 2022 to April 2023 demonstrates that the RF model trained on GPS data achieves a high coefficient of determination (r) of 0.98 for the training set, accompanied by a mean absolute error (MAE) of 0.021 m. For the test set, r is 0.95, and the MAE is 0.033 m. The RF model trained on BDS data exhibits a coefficient r of 0.98, with a corresponding MAE of 0.024 m. For the test set, r is 0.94, and the MAE is 0.038 m. These results underscore the efficacy of machine learning methodologies and the benefits of integrating FY-3E data with SMOS sea ice parameters for SIT estimation.