Sea Ice Characterization from Earth Observation Data by Explainable Artificial Intelligence

The project will concentrate on estimation of the most essential sea ice (SI) parameters and SI classification from space-borne microwave earth-observation (EO) instruments and ice models by deep machine learning (ML). There exist notable deficiencies in the current ML SI algorithms. Typically, ML appears as a black box: no or only little information on the relationship between the ML inputs, structure of the ML, representations encoded by the ML, and the outputs are available. Also, no local uncertainties of the ML estimates are provided. These deficiencies will be addressed by developing an integrated set of optimized sea ice classification and parameter estimation algorithms with involved local uncertainties. Explainable AI (XAI), including sensitivity analysis, will be applied for analyzing and developing the ML. ML hyper-parameter (structure) optimization will also be applied to optimize the algorithms. The proposed work will be performed using existing Baltic Sea and Arctic data.