We have developed a process for evaluating future satellite-based hyperspectral microwave sensor designs using deep neural networks (DNN). Our approach combines a sophisticated simulated data product with a hierarchical deep neural network capable of comparing the relative performance of a variety of different microwave sounder configurations. These configurations include both spectral band coverage and resolution which allows for a thorough investigation of the solution space. The relative performance between these configurations as tested on the prediction of the planetary boundary layer height (PBLH) is used to perform the evaluation. We plan to extend this method to the prediction of entire temperature and water profiles to further refine this process.