Variational retrieval of rain profiles from spaceborne passive microwave radiance observations

A variational retrieval method has been implemented to evaluate the possibility of assimilating rain-affected microwave radiances in an operational weather forecasting model. The method employs an efficient multiple scattering radiative transfer code and its tangent-linear and adjoint versions to simulate model state equivalent radiances and their sensitivities to changes in hydrometeor contents. Profiles of temperature and humidity together with the ECMWF model convection and cloud schemes are used to produce first-guess hydrometeor profiles. The hydrometeor error covariance matrix is calculated from the operational error covariance matrix of temperature and humidity convolved with both convection and cloud scheme for each profile. A sensitivity study is applied to a tropical cyclone using data from the TRMM Microwave Imager where large differences between the first-guess and observed precipitation fields occurred. The study shows that the approach results in single-column retrievals that compare well with an independent algorithm. A strong dependence of the analysis on the microwave frequencies chosen as observations was noted. This was especially the case in areas where some microwave channels lost their sensitivity to modifications of the profiles due to saturation or where ambiguous contributions from either cloud droplet absorption or raindrop absorption and scattering were present. Including all channels at 10.7, 19.35, 21.3, and 37.0 GHz produced the lowest number of minimization failures. The failures are either due to strongly non-linear sensitivities of radiances to changes in hydrometeor contents for a given profile or due to areas where the channel dependent spatial resolution produces inconsistent observations.