scieee AI-readable full text Open interactive document viewer

2019-2020 Arctic sea ice lead frequency distribution dataset

Yang, Xinran; Yibin, Ren; Li, Xiaofeng

Abstract

The 2019-2020 Arctic sea ice lead frequency distribution dataset provides a comprehensive, high-resolution record of sea-ice lead distribution across the pan-Arctic region from October 2019 to April 2020. It includes 14,093 pixel-level binary lead classification maps derived from Sentinel-1 EW SAR imagery, together with seven monthly lead-frequency composites that quantify the spatial and temporal variability of leads on an 80 m × 80 m grid north of 65° N. All products are provided as georeferenced GeoTIFF files using the WGS 84 coordinate system. The dataset captures both scene-level lead morphology and monthly lead occurrence patterns, enabling large-scale analysis of sea-ice dynamics, thermodynamic exchange, and atmosphere–ocean interaction. All data are packaged in ZIP archives for efficient distribution, with LeadNet_results organized into seven monthly ZIP files. A supplementary code package (code.zip) includes the training and inference scripts used in the LeadNet workflow and tools for computing lead-frequency maps. This dataset serves as a reusable benchmark for machine-learning-based lead detection and Arctic climate research.

Full text

2019-2020 Arctic sea ice lead frequency distribution dataset The LeadNet Arctic Lead Dataset consists of two complementary data products: (i) Monthly pan-Arctic lead-frequency (LF) maps and (ii) Pixel-level binary lead classifications for individual Sentinel-1 SAR scenes. All data are provided as GeoTIFF files on a regular 80 m × 80 m grid north of 65° N, using the WGS 84 geographic coordinate system (EPSG:4326). The dataset covers 14,093 Sentinel-1 EW scenes acquired from October 2019 to April 2020, yielding seven monthly LF files and 14,093 LeadNet classification files, totaling approximately 1.09 TB. All files have been packaged as ZIP archives for convenient download. - Monthly LF maps are stored in a single ZIP file. - LeadNet_results are divided into 7 ZIP packages following the naming format LeadNet_results_YYYYMM.zip. - An additional archive, code.zip, provides all scripts used in the associated publication, including: ¡ LeadNet_Train_and_Test.ipynb – training and testing code used in the article; ¡ Lead_test.py – batch-processing prediction script using the trained model LeadNet.h5 to generate LeadNet_results; ¡ LF_calculate.py – the script computing monthly LF maps from LeadNet_results. The full dataset is publicly available at DOI: 10.5281/zenodo.17718323, and serves as a reusable benchmark for Arctic seaice lead detection and surface heat-flux studies. 1. Monthly lead frequency maps Each monthly file (LeadFreq_YYYYMM.tif) contains a single band representing lead frequency as a percentage (%), with valid values from 0 to 100. - The data format used is GeoTIFF. The GeoTIFF format preserves complete georeferencing information (WGS 84) and is compatible with mainstream GIS and remote sensing software. - The file structure is as follows: ¡ All files follow the naming convention LeadFreq_YYYYMM.tif, where YYYYMM represents the year and month of the composite. ¡ Each file covers the pan-Arctic region north of 65° N. ¡ The spatial resolution is 80 m × 80 m, and the image extent varies slightly among months. ¡ Each pixel value represents the monthly occurrence frequency (0–100 %) of open leads. ¡ Metadata includes projection information, geographic extent, resolution, and processing date. 2. LeadNet classification results Each file (LeadNet_YYYYMMDD_SceneID.tif) contains one band with integer values, where 1 denotes background (sea ice and open ocean), 2 denotes open-water leads, and 0 is used as the no-data value. The characteristics of the LeadNet detection results are listed below. - The data format used is GeoTIFF, containing the LeadNet segmentation results for each available Sentinel-1 scene. - The file structure is as follows: ¡ All files follow the naming convention LeadNet_YYYYMMDD_SceneID.tif, where YYYYMMDD indicates the acquisition date of the Sentinel-1 image, and SceneID represents the unique ID of the original Sentinel-1 image. ¡ Each file contains a single-band binary image with pixel values of 1 (background) and 2 (lead). ¡ Spatial resolution and coordinate reference are consistent with the original Sentinel-1 input data. ¡ Metadata includes projection information, geographic extent, resolution, and processing date.