Thin Sea Ice Thickness Prediction Using C-band Polarimetric Radar and Physical Observations Via Multivariate Random Forest Algorithm

This study improves understanding of the prediction of Arctic thin sea ice thickness by incorporating 11 physical and radar parameters into a multivariate Random Forest algorithm. The results demonstrated a highly accurate prediction with RMSE of 0.01 cm for thin ice layer and a low Out-of-Bag error validating the training process. Using the Permutation Importance method, the study ranked Cumulative Freezing Degree Moments, various salinity measurements, and radar Cband backscattering values (in VV and HH polarizations) as the most crucial parameters in the prediction. These findings can guide future research toward the feasibility of sea ice thickness prediction and improving efficiency in data preparation and processing.