Arctic Sea-ice Type Recognition Based on the Surface Wave Investigation and Monitoring Instrument of the China-French Ocean Satellite

Abstract

The sea-ice type is the key parameter for Arctic sea-ice monitoring. Traditionally, Arctic sea-ice types are mainly identified based on the scatterometer, radiometer, and Synthetic Aperture Radar (SAR) with the medium-incidence angles (about 20°–60°). The Surface Wave Investigation and Monitoring instrument (SWIM) on the China-France Oceanography Satellite (CFOSAT), as a new type sensor with the low-incidence angles (0°, 2°, 4°, 8°, and 10°), is different from traditional remote sensors. SWIM has the potential to recognize sea-ice types based on its echo signals and backscattering information, whereas relevant research is still under development. In this research, the sea-ice classification using the SWIM’s observations of the low-incidence angles is studied. Firstly, the waveform features of the six low-incidence angles were extracted from the SWIM echoes from November 2019 to April 2020, including the maximum power (MAX), the backscattering power (BSP), the pulse peakiness (PP), the stack standard deviation (SSD), the leading-edge width (LEW), and the trailing-edge width (TEW). Secondly, the Euclidean distances among the sea-ice types (the first-year ice (FYI) and the multi-year ice (MYI)) and the open water (OW) are used to analyze the distinction of the three categories based on the waveform features in the incidence angles. Thirdly, the FYI, MYI, and OW are classified by the Support Vector Machine (SVM) using the waveform features, and the three kinds of kernel (Gaussian kernel, Linear kernel, and Polynomial kernel) for SVM are analyzed for their capabilities of sea-ice classification. Then, the classification results are compared with the ice charts of the Arctic and Antarctic Research Institute (AARI) to evaluate the classification accuracies. It is found that the accuracies of the waveform features in the low-incidence angles for the Gaussian kernel are the highest, and the accuracies of the Linear kernel are the lowest. It is revealed that the accuracies of all features in the different angles using Gaussian kernel is up to 77%. It is concluded that SWIM has the capacity for the sea-ice classification. In the future, our work will focus on the development of recognition methods based on more waveform features or feature sets to improve the accuracies of the sea-ice classification. This research will expand new horizons for the SWIM ocean detection application.