Satellite observations have become an instrumental component of air pollution exposure estimation for use in health effects studies, providing critical spatial and temporal information beyond what can be provided from ground-based monitoring networks. The aerosol retrieval for the upcoming MultiAngle Imager for Aerosols (MAIA) instrument involves an optimization algorithm that uses observed multiangular, multispectral, and polarimetric radiances to generate a set of estimated aerosol characteristics including aerosol optical depths (total and by size and type). We demonstrate a testbed that includes several statistical and machine learning methods that pinpoint the most reflective set of MAIA retrieved optical properties that robustly estimate component-specific air pollution concentrations. As part of this testbed we also address the many sources of uncertainty that contribute to the overall retrieval uncertainty, including measurement uncertainty, uncertainties in initializations of the algorithm, and uncertainties due to the forward model. Our results provide critical preparation for the launch of MAIA.