Oral

IACS

JCP07 - Remote Sensing of Sea Ice from Satellite Microwave Measurements and its Applications

Reference sea ice concentration data records from Landsat-8 imagery and its applications

Since satellite-borne passive microwave (PMW) observation has been available, Arctic sea ice coverage, commonly represented as sea ice concentration (SIC), has been continuously monitored due to the extensive spatiotemporal coverage and all-weather capability of PMW instruments. However, advancements in SIC estimation algorithms have been constrained by the absence and/or limited of reliable reference SIC datasets, leading to inconsistencies among existing passive microwave-based SIC products. In order to address this kind of issue, a reference SIC dataset spanning three years was generated using the measured high-resolution images from the Operational Land Imager (OLI) onboard Landsat-8. Given the significant uncertainties associated with cloud masking over Arctic sea ice, manual inspection of true-color images is conducted to categorize each Landsat-8-measured scene into four groups based on cloud mask accuracy. Sea ice is distinguished from open water using the Normalized Difference Snow Index (NDSI) and OLI band-5 reflectance. The classified data are then reprojected onto a 6.25 km polar stereographic grid, with SIC determined by the ratio of the number of sea ice-classified pixels and the number of total pixels within each grid cell. From a total of ~15,000 Landsat-8 scenes, nearly 93% of the scenes were successfully converted into SIC maps, yielding nearly 3 million SIC grid cells. The generated SIC dataset is validated against ice chart-derived SICs, demonstrating a strong linear relationship and confirming its potential as a reference dataset for evaluating and improving existing SIC products. A detailed description of the dataset is provided by Jung et al. (2024), and the dataset is publicly available at https://zenodo.org/doi/10.5281/zenodo.10973297. Using this reference SIC dataset, the study evaluates SIC products derived from the NASA Team (NT) and Bootstrap (BT) algorithms, applied to Special Sensor Microwave Imager and Sounder (SSMIS) data archived at the National Snow and Ice Data Center (NSDIC). Overall, BT-based SICs show better agreement with the reference SICs across seasons, particularly at higher SIC values. However, both algorithms struggle to differentiate ice from open water in low SIC regions. In addition to evaluating the existing SIC algorithms, the constructed dataset can be used for developing a machine-learning algorithm for estimating SICs over the Arctic. In this study, a multilayer perceptron model for such a purpose has been developed and the results will be shown.