Advancing Sea Ice Classification Capabilities in SAR Imagery Via Polarimetric Analysis and Machine Learning

Understanding rapid changes in sea ice is essential for Arctic navigation. While machine learning models derived from remote sensing imagery are ideal for this work, they are limited by a current lack of high-fidelity training label datasets. To address this, we are developing methods for deriving labels directly from high-fidelity synthetic aperture radar data using polarimetric and texture feature analysis. We use decisionbased supervised classifiers to identify optimal polarimetric and texture feature relationships for class separation and subsequently use these relationships to establish new classification planes optimized for sea ice identification in SAR data. We also establish methods to quantify uncertainty in the resulting labels. These theoretical, physics-based planes improve the standard of label resolution tenfold, are robust to confounding factors such as wind-roughened waters and inaccurate training labels, and have the potential to be satellite-agnostic, thereby advancing the current state of sea ice label derivation.