Rainfall Microwave Spectral Atoms: A New Class of Bayesian Algorithms for Passive Retrieval

Abstract:

To improve the quality of precipitation retrievals from space, especially over land, it is important to increase our understanding of dominant precipitation microwave spectral signatures and their corresponding rainfall profiles. In this presentation, we demonstrate that despite markedly variable space-time structure of precipitation processes, there are a few elementary microwave spectral atoms that can be learned and exploited for compact and sparse retrieval of rainfall profiles from their spectral signatures in microwave bands between 10-183 GHz. For different land surface radiation regimes, these atoms are encoded via supervised learning of two discriminative dictionaries from a large set of wet and dry spectral signatures obtained from coincident measurements of precipitation radar and radiometer on board the Global Precipitation Measuring (GPM) satellite. A new Bayesian rainfall retrieval algorithm is presented to detect and reconstruct the precipitation profiles using linear combinations of sparse groups of the microwave spectral dictionary atoms.