This dissertation explores machine learning approaches for estimating the refractivity within the marine atmospheric boundary layer (MABL) under various electromagnetic ducting conditions. We use simulated radar propagation data that is representative of data that can be sparsely measured in practice. In conjunction with the sparse data collection scheme, a trained artificial neural network can be used to effectively characterize evaporation duct height (EDH) from the data, in real-time. We further show that Gaussian process regression (GPR) can accomplish this task, and produce uncertainty quantification on the EDH predictions, also in real-time. Finally, we show that a two-step deep learning model can classify and characterize different types of ducting conditions.