Reconstructing daily snow and ice albedo series for Greenland by coupling spatiotemporal and physics-informed models

• Published in: International Journal of Applied Earth Observation and Geoinformation
• Main Authors: Fan Ye, Qing Cheng, Weifeng Hao, Dayu Yu, Chao Ma, Dong Liang, Huanfeng Shen
• Format: Article in Journal/Newspaper
• Language: English
• Published: Elsevier 2023
• Subjects: Albedo; Gap-filling; Snow and ice; Greenland; Physical geography; GB3-5030; Environmental sciences; GE1-350; Ice Sheet
• Online Access: https://doi.org/10.1016/j.jag.2023.103519; https://doaj.org/article/6694c2e30bbb4fbd91c63f9ae714d56f

Snow and ice albedo is a critical geographical indicator that reflects climate change on Earth. Quantifying the albedo in Greenland ice sheet, which is extensively covered with snow and ice, is key to studying changes in the energy budget in the Northern Hemisphere. Earth observation satellites have been regularly providing surface albedo products. However, optical satellite-derived albedo products have many voids due to the persistent cloud cover over the Greenland ice sheet. Consequently, seamless reconstruction of albedo on the spatial and temporal scales is essential. Surface albedo, as a geographical element, is spatially and temporally correlated. In addition, the broadband albedo of snow and ice is significantly modified by changes in the spectral distribution of solar irradiance caused by clouds. On the basis of such facts, this study proposes a reconstruction method for snow and ice albedo that combines spatiotemporal information with a physics-informed model. This method uses spatiotemporal nonlocal filtering to generate the initial reference albedo for missing pixels. Then, the hypothetical clear-sky albedo is reconstructed using the Whittaker iterator. Finally, cloudy albedo is obtained on the basis of the empirical relationship between clear-sky and cloudy-sky albedos. We reconstruct albedo based on MOD10A1 for the whole Greenland region from 2001 to 2020. A comparison between the reconstruction results and ground measurements exhibits satisfactory accuracy with an R-value of 0.8162, a root-mean-square error of 0.0669, a mean absolute error of 0.0486, and a bias of 0.00001. Moreover, the proposed method demonstrates the advantages of being more accurate and robust than other classical methods. Therefore, this study will be valuable for generating 500 m daily remotely sensed albedo of snow and ice in large regions.