Sentinel-1 Sea Ice Concentration Products for the Melting Season in Fram Strait, Arctic (2021–2023)

The 1km resolution Sea Ice Concentration (SIC) dataset for Fram Strait in the Arctic during the melting periods (June to September) from 2021 to 2023 is generated using the Ice-WaterNet framework. To address the challenges posed by complex surface features during the melt season, such as wind-induced roughness in open-water regions and varying ice floe sizes, a superpixel-based approach is implemented. This approach effectively minimizes the uncertainty of ice-water classification by integrating Conditional Random Fields (CRF) with U-Net models. CRF is used to identify uncertain superpixels, while uncertainty is progressively reduced through iterative refinement within the edges of superpixels, utilizing the U-Net attention mechanism for enhanced accuracy. The dataset includes daily SIC products available in both NetCDF and GeoTIFF formats, along with the corresponding georeference information provided in NetCDF format. Additionally, a separate textfile detailing the color indexing is included for visualization purposes.

2021-2023年北极弗拉姆海峡（Fram Strait）融化期（6-9月）的1千米分辨率海冰浓度（Sea Ice Concentration, SIC）数据集由Ice-WaterNet框架生成。为应对融化季复杂表面特征带来的挑战（如开阔水域的风致粗糙度及浮冰（ice floe）大小的变化），本研究采用了基于超像素的方法。该方法通过将条件随机场（Conditional Random Fields, CRF）与U-Net模型相结合，有效降低了冰水分类的不确定性。CRF用于识别不确定的超像素，同时利用U-Net注意力机制在超像素边缘内进行迭代细化，逐步降低不确定性以提高精度。数据集包含每日SIC产品，提供NetCDF和GeoTIFF两种格式，同时附有NetCDF格式的对应地理参考信息。此外，还包含一个独立的文本文件，详细说明用于可视化的颜色索引。