Melt pond fractions on Arctic summer sea ice retrieved from Sentinel-3 satellite data with a constrained physical forward model

• Published in: The Cryosphere
• Main Authors: Niehaus, Hannah, Istomina, Larysa, Nicolaus, Marcel, Tao, Ran, Malinka, Aleksey, Zege, Eleonora, Spreen, Gunnar
• Format: Article in Journal/Newspaper
• Language: English
• Published: Copernicus Publications 2024
• Subjects: article; Verlagsveröffentlichung; Arctic; albedo; Sea ice; The Cryosphere
• Online Access: https://doi.org/10.5194/tc-18-933-2024; https://noa.gwlb.de/receive/cop_mods_00072037; https://noa.gwlb.de/servlets/MCRFileNodeServlet/cop_derivate_00070267/tc-18-933-2024.pdf; https://tc.copernicus.org/articles/18/933/2024/tc-18-933-2024.pdf

The presence of melt ponds on Arctic summer sea ice significantly alters its albedo and thereby the surface energy budget and mass balance. Large-scale observations of melt pond coverage and sea ice albedo are crucial to investigate the role of sea ice for Arctic amplification and its representation in global climate models. We present the new Melt Pond Detection 2 (MPD2) algorithm, which retrieves melt pond, sea ice, and open-ocean fractions as well as surface albedo from Sentinel-3 visible and near-infrared reflectances. In contrast to most other algorithms, our method uses neither fixed values for the spectral albedo of the surface constituents nor an artificial neural network. Instead, it aims for a fully physical representation of the reflective properties of the surface constituents based on their optical characteristics. The state vector X, containing the optical properties of melt ponds and sea ice along with the area fractions of melt ponds and open ocean, is optimized in an iterative procedure to match the measured reflectances and describe the surface state. A major problem in unmixing a compound pixel is that a mixture of half open water and half bright ice cannot be distinguished from a homogeneous pixel of darker ice. In order to overcome this, we suggest constraining the retrieval with a priori information. Initial values and constraint of the surface fractions are derived with an empirical retrieval which uses the same spectral reflectances as implemented in the physical retrieval. The snow grain size and optical thickness change with time, and thus the ice surface albedo changes throughout the season. Therefore, field observations of spectral albedo are used to develop a parameterization of the sea ice optical properties as a function of the temperature history of the sea ice. With these a priori data, the iterative optimization is initialized and constrained, resulting in a retrieval uncertainty of below 8 % for melt pond and 9 % for open-ocean fractions compared to the reference dataset. As ...