An observationally based method for stratifying a priori passive microwave observations in a Bayesian‐based precipitation retrieval framework

Abstract

Estimation of precipitation from space‐based passive microwave (PMW) radiometric brightness temperature (TB) observations that adapts to the wide variety of Earth surface background and environmental conditions is a long‐standing issue. Since these conditions are generally unknown from the TB observations, PMW‐based precipitation estimation techniques commonly utilize independent ancillary data sources, such as interpolated prognostic variables from numerical weather prediction forecast models, and discrete surface emissivity classifications. In some situations, the selection of these variables may restrain the algorithm performance under particular surface and atmospheric conditions. The objective of this article is to examine the emissivity principal component (EPC) analysis as a common stratification method for indexing, searching and weighting candidate precipitation profiles from a priori databases, adaptable for Bayesian‐based precipitation estimation algorithms applied to the Global Precipitation Measurement (GPM) Microwave Imager (GMI) or other PMW sensors, to minimize dependence upon ancillary data sources. The EPC has been previously shown to track the joint variability between the 10–89 GHz surface emissivity, total column precipitable water vapour (TPW) and surface temperature (Ts) conditions directly from the TB observations, and identify global locations of similar conditions. A parallel GMI precipitation retrieval was carried out where the identical a priori database was indexed by TPW, Ts and a surface emissivity class index. An independent validation of each precipitation retrieval scheme was carried out using GMI pixel‐matched Multi‐Radar Multi‐Source (MRMS) ground radar data over the continental USA and surrounding ocean waters. While the EPC‐based estimates demonstrated similar performance to the TPW‐based estimates over ocean backgrounds, a markedly improved detection, and reduction in bias, was found for moderate and higher (>5 mm/hr) rainfall rates over other backgrounds, especially vegetated surfaces and coastlines.