Performance Analyses of Passive Microwave Atmospheric Sounding Approaches That Use Hyper-Spectral-Sampling and/or Multi-Angle-Sampling: Methods, Simulation Examples, and Pitfalls

In this paper, the key models, assumptions, and statistical characterizations/uncertainties are examined to better understand how to evaluate hyper-spectral sampling (HSS) and "multi-angle-sampling (MAS) systems and interpret results that are commonly presented in the contemporary literature. The analysis presented is based on a set of thousands of representative, global atmospheric profiles over a variety of surfaces. These profiles are used with a line-by-line radiative transfer model to calculate the at-sensor radiance for a variety of viewing angles, spectral response functions, channelization schemes, and noise assumptions. A neural-network-based retrieval scheme [Blackwell, 2005] similar to one that is used for AIRS/AMSU on Aqua and for TROPICS near-real-time processing is used to retrieve the profiles. Errors are characterized by bias, RMS uncertainty, averaging kernel width, and retrieved profile error correlation width for a variety of sensor configurations to explore HSS and MAS performance. Neural network jacobians [Blackwell, 2012] facilitate the direct calculation of these metrics in some cases.