Retrieval of thin ice thickness from multifrequency polarimetric SAR data

We discuss the observed C- and L-band polarimetric signatures of thin lead ice in one Synthetic Aperture Radar (SAR) image based on the expected ice properties and results from a scattering model. In this article, we focus on thin ice with thicknesses in the range of 0–10 cm. The layered scattering model used here allows for the inclusion of surface and volume scattering contributions from a slush layer, an ice layer, and roughness at the interfaces. The sensitivity of the signatures to the model parameters is explored. A highly saline surface skim formed on the top surface during ice growth significantly affects the electromagnetic properties of the medium and helps to explain the magnitude of the copolarized returns at high incidence angles. Based on these model predictions, we demonstrate an approach to retrieve the ice thickness from polarimetric SAR observations. The approach includes the training of a neural network with model predictions and using this neural network to estimate the ice thickness distribution using polarimetric observations from SAR data. The results from this ice thickness retrieval process are discussed.