On adapting data assimilation framework to data fusion of multi-scale precipitation observations

Data assimilation (DA) based on estimation theories has been a powerful tool to extract information from a wide range of data sources to produce optimal estimates of physical parameters. In operational NWP applications, the available information essentially consists of observations, the physical laws governing the dynamical and physical processes of the atmosphere, as well as the associated uncertainties of these information sources. The DA framework can be adapted to the multi-sensor multi-scale data fusion problem, in which the primary goal is not to define the initial condition of a forecast as in operational DA systems, but to produce an estimate of the physical field of interest as accurate as possible. In this work we explore the DA framework in an application of combining multi-sensor multi-scale precipitation data to produce an integrated observation-based precipitation analysis. Similar to a DA system, an objective function is optimized with minimization of analysis error variance, provided that the error characteristics for each data source are estimated and described by its error covariance. Wavelet-transform is employed to obtain scale-decomposed and sparse representation of signal distribution. We present a prototype data fusion system including the estimation/fusion algorithm based on data assimilation methodologies, the estimation of uncertainty variance database for satellite retrievals, and the data registration for different platforms. A set of case studies such as hurricanes and mesoscale convective complexes are used to evaluate the general paradigm of the data fusion system and to investigate the impact from individual data source, different spatial resolutions and observation error distributions.