Quantification of the Effects of Preprocessing Filters on the Performance of GNSS-R Based Sea Ice Detection

This study examines the influence of preprocessing techniques on the performance of Global Navigation Satellite System Reflectometry (GNSS-R) based sea ice detection. Preprocessing techniques evaluated in this study include precipitation screening and SNR thresholding. SNR thresholds are tested with −3dB, 0 dB, +3 dB, and no threshold configurations. Precipitation screening is tested in three scenarios, containing all data, precipitation free data, and only precipitation contaminated data. Performance impact is determined by recording the sea ice detection accuracy, precision, and recall using the preprocessed datasets. The binary classification is conducted using two models: a histogram thresholding approach and a Bayesian approach. From these tests it was found that the inclusion of precipitation contaminated data did not result in an appreciable change to classification accuracy, precision, or recall. For SNR thresholding, classifcation accuracy peaks for the histogram thresholding approach when a SNR threshold of −3dB or 0 dB is applied. For the Bayesian approach, peak accuracy is achieved with a −3 dB threshold. These results indicate that an SNR threshold of < 0 dB results in superior classification accuracy and that precipitation may be considered negligible for sea ice detection purposes.