An Evaluation of Ground Truth Data on GNSS-R Based Sea Ice Classification

Previous studies about Global Navigation Satellite System Reflectometry (GNSS-R) based Arctic sea ice classification use either the Ocean and Sea Ice Satellite Application Facility (OSI SAF) ice type dataset or sea ice charts produced by the U.S. National Ice Center (USNIC). These data differ in the labels provided and how they are derived from the original satellite readings. This study aims to explore the impact of the selected ground truth data on the performance of GNSS-R based sea ice classification. This is done by training random forest (RF) classifiers using the same classification features, but two different ground truth datasets. The RF model trained on the NSIDC dataset yielded improved performance metrics, with an increased Kappa coefficient of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{0. 6 8 9 8}$</tex>, indicating much greater agreement between predictions and ground truth.