Machine-learning-based classification and spatiotemporal analysis of sea ice in the Bohai Sea

In recent years, frequent sea ice disasters have led to severe disruptions in marine ecology and hindered maritime transportation. In this study, Sentinel-1 data from 2016 to 2022 were utilized to extract texture features from images. A comparative analysis of the classification accuracy of three models—Support Vector Machine, Iterative Self-Organizing Clustering, and Random Forest—was conducted. A high-precision sea ice classification model was established to analyze spatiotemporal changes. The results indicate that the Support Vector Machine model exhibited the highest accuracy, with an overall accuracy of 87.61% and a kappa coefficient of 81.42%, demonstrating the model's stability.