Satellite-Estimated Microwave Emissivity and Emission Temperature Over the Arctic Sea Ice: ANN-Based Algorithm

Satellite microwave surface-sensitive sounding channels have long faced limitations in polar data assimilation (DA) due to difficulties in specifying radiative boundary conditions over sea ice, particularly during winter. Previous studies have proposed using outputs from sea-ice growth model to estimate microwave emissivities and emission temperatures for DA. However, this approach poses practical challenges, as it operates independently of numerical weather prediction systems. To overcome these challenges, this study introduces an artificial neural network (ANN) technique that transforms the modeling approach into an algorithm capable of directly retrieving radiative boundary conditions from satellite observations. The ANN leverages empirical knowledge that sea-ice emissivity is closely related to ice type and that microwave penetration depth varies with frequency, capturing temperature profiles within sea ice. It targets winter Arctic Ocean, using brightness temperatures (TBs) from ATMS channels 1 to 3 (23, 36, and 51 GHz) as inputs to retrieve emissivity and emission temperature for surface-sensitive channels 6 and 7 (53 and 54 GHz). The ANN algorithm was trained and validated using simulated data sets, demonstrating high accuracy with correlation coefficients of 0.99 and low errors. Further evaluation using ATMS observations from the winters of 2021 and 2022 was conducted by converting the retrievals into top-of-atmosphere TBs, yielding consistent error statistics: correlation coefficients of 0.99, mean deviations of about −0.3 K, and standard deviations of about 0.3 K for both channels. These results confirm the algorithm's stability and robust generalization capabilities, underscoring its potential for integration within DA systems and broader applications to other microwave sensors.