Over-ocean rainfall retrieval from multisensor data of the tropical rainfall measuring mission. Part I: Design and evaluation of inversion databases

A new methodology for the combination of active and passive microwave measu rements for near-surface precipitation retrieval from the Tropical Rainfall Measuring Mission (TRMM) data was developed. The approach consists of a st and-alone passive microwave algorithm that is calibrated by collocated rada r estimates. The passive microwave technique was based on combined cloud mo del-radiative transfer simulations including varying surface conditions, a melting layer parameterization, and approximative three-dimensional radiati ve transfer. The representativeness of the simulations with respect to the TRMM Microwave Imager (TMI) observations was evaluated replacing brightness temperatures by empirical orthogonal functions. Thus, nine TMI correlated channels may be replaced by two to three empirical orthogonal functions rep resentating 97%-98% of total variability. Comparing the principal component s to those from TMI observations containing precipitation revealed that the 85.5-GHz brightness temperatures from the simulations represent the major source of mismatch. This is due to the accumulation of uncertainties in clo ud model parameterizations of ice microphysics and approximative radiative transfer at this frequency where scattering is most efficient. Depending on the lowest detectable rainfall threshold, the simulations covered 88%-99% of observations from collocated TMI-precipitation radar measurements. Gaps occurred mostly for less intense cloud systems that are not well represente d by the cloud model simulations. The ambiguity of observations, that is, t he multiplicity of hydrometeor profiles with the same passive microwave sig nature, was also analyzed. It was found that ambiguity decreases with incre asing intensity of the observed scene. In terms of near-surface rain liquid water content, the standard deviation reaches 50%-100% for less intense ra in (0.01 g m(-3)) and is reduced to 20%-30% for intense rain (1.0 g m(-3)) events. Excluding the 85.5-GHz channels clearly produced less ambiguity. Ab out 80%-95% of all cases showed less than 50% standard deviation of the ret rieval variable per database entry compared to 65%-85% when the 85.5-GHz ch annels were included.