Neural-Network based algorithm for ice concentration retrievals from satellite passive microwave data

Present algorithms for observing the multiyear ice cover are not accurate in multiyear fraction calculations, which is a significant disadvantage of the present system of global ice monitoring considering the fact that multiyear ice is one of the key indicators of changes in the Arctic climate. In this research regionally differing Neural Networks (NN)-based algorithms for total and multiyear Arctic sea ice concentration retrievals from Special Sensor Microwave Imager (SSM/I) data are developed using closed scheme of the numerical experiment. Era-40 Reanalysis data on atmospheric parameter profiles and sea ice temperature are used for the numerical integration of the radiation transfer of the microwave emission in the Atmosphere-Ocean-Ice System. The data on cloud liquid water content and cloud boundaries are modeled basing on the results of Arctic SHEBA experiment. Numerical values for first year and multiyear ice emissivities are taken from published experimental data. The calculated radiometer brightness temperature values are used for NN-based theoretical algorithm development. New weather filter is defined. The algorithms are validated for stable winter conditions using collocated SSM/I data and Synthetic Aperture Radar (SAR) images, classified by an ice expert.