GNSS-R Based Sea Ice Classification Using Track Normalized Observables

This study investigates per track observable normalization for Global Navigation Satellite System Reflectometry (GNSS-R) based sea ice classification. The objective of this technique is to compensate for reflected power inconsistencies retrieved from the TechDemoSat-1 (TDS-1) GNSS-R satellite. Classification features include delay Doppler map average (DDMA) and per track Z-score normalized DDMAs to compensate for the TDS-1 calibration. Ground truth labels are derived using the National Snow and Ice Data Center (NSIDC) sea ice concentration dataset. The random forest (RF) classifier yielded recall values 91.07%, 60.29%, and 81.22% for first-year, multi-year, and thin ice respectively. The results indicate that the per track normalization improves performance relative to previous literature that classified sea ice in three different types.