DataSheet_1_Sea ice recognition for CFOSAT SWIM at multiple small incidence angles in the Arctic.docx

• Main Authors: Meijie Liu, Ran Yan, Xi Zhang, Ying Xu, Ping Chen, Yongsen Zhao, Yuexiang Guo, Yangeng Chen, Xiaohan Zhang, Shengxu Li
• Format: Dataset
• Language: unknown
• Published: 2022
• Subjects: Oceanography; Marine Biology; Marine Geoscience; Biological Oceanography; Chemical Oceanography; Physical Oceanography; Marine Engineering; sea ice; Surface Wave Investigation and Monitoring (SWIM); Arctic; small incidence angles; waveform features; k-nearest neighbor method; Sea ice
• Online Access: https://doi.org/10.3389/fmars.2022.986228.s001; https://figshare.com/articles/dataset/DataSheet_1_Sea_ice_recognition_for_CFOSAT_SWIM_at_multiple_small_incidence_angles_in_the_Arctic_docx/21086656

Sea ice recognition is one of the main tasks for sea ice monitoring in the Arctic and is also applied for the detection of other ocean phenomena. The Surface Wave Investigation and Monitoring (SWIM) instrument, as an innovative remote sensor that operates at multiple small incidence angles, is different from existing sensors with moderate and normal incidence modes for sea ice monitoring. Sea ice recognition at small incidence angles has rarely been studied. Moreover, SWIM uses a discrimination flag of sea ice and sea water to remove sea ice from sea wave products. Therefore, this research focuses on sea ice recognition in the Arctic based on SWIM data from October 2020 to April 2021. Eleven features are first extracted, and applied for the analysis of the waveform characteristics using the cumulative probability distribution (CPD) and mutual information measurement (MIM). Then, random forest (RF), k-nearest neighbor (KNN) and support vector machine (SVM) classifiers are built, and their abilities of sea ice recognition are assessed. The optimal classifier is the KNN method with Euclidean distance and k equal to 11. Feature combinations are also used to separate sea ice and sea water based on the KNN method to select the optimal combination. Thus, the optimal classifier-feature assembly at each small incidence angle is established, and the highest overall accuracy reaches 97.1%. Moreover, the application of the optimal classifier–feature assemblies is studied, and its performance is fairly good. These assemblies yield high accuracies in the short- and long-term periods of sea ice recognition, and the overall accuracies are greater than 93.1%. So, the proposed method satisfies the SWIM requirement of removing the sea ice effect. Moreover, sea ice extents and edges can be extracted from SWIM sea ice recognition results at a high level of precision greater than 94.8%. As a result, the optimal classifier–feature assemblies based on SWIM data express the effectiveness of the SWIM approach in sea ice recognition. Our ...