Implementation and Investigation on the Assimilation of Precipitation-Affected Microwave Radiances in the NCEP FV3DA System

The current NCEP Grid-point Statistical Interpolation (GSI) analysis system has been utilizing hybrid 4D ensemble-variational scheme for data assimilation since 2016. The satellite radiance data from AMSU-A are assimilated under the all-sky condition in a way that only the clear radiances and radiances affected by non-precipitating clouds (liquid water and ice clouds) are assimilated, and this is because only the non-precipitating clouds are prognostics in the forecast model and the precipitation amount is diagnostics at the surface only. In early 2019, the current forecast model will be replaced with the Finite-Volume version 3 (FV3) based Next Generation Global Prediction System (NGGPS) model in which five hydrometeors types (cloud water, ice, rain, snow, and graupel) are prognostic variables along with diagnostic cloud fraction at each model level.

In preparing GSI for assimilating the full set of FV3 hydrometeors, the Community Radiative Transfer Model (CRTM) has been validated and improved its accuracy under the multiple-scattering condition which involves precipitation (rain and snow) in microwave spectral range. Moreover, a two-column radiance calculation in CRTM was developed to more realistically handle cloudy scenes with fractional cloudiness. The current CRTM can only handle overcast condition, and this could introduce biases up to 100 K in small-scale convective cloudy regions for high- frequency microwave channels. A cloud optical table constructed using the particle size dependent mixture of 6 distinct type of ice crystal habit is included in the CRTM for cloudy simulation.

The control and state variables, as well as the ensemble perturbations in GSI analysis system, have been augmented to include the precipitating hydrometeors (rain, snow, and graupel) on top of the existing non-precipitating types (cloud liquid water and ice). The ensemble perturbations provide flow-dependent background error along with the static background error covariance for each hydrometeor which is set to 5% of its background value. The correlation lengths for hydrometeors are set to 50% of those assigned to moisture. The data thinning and quality control are relaxed to include radiances with precipitation, and the observation error for each AMSU-A channels are re-estimated to include the data affected by precipitation as the function of cloud amount.

The implementation related to assimilating precipitation-affected microwave radiances in both GSI and CRTM will be described in detail. The impact of assimilating precipitation-affected AMSU-A radiance on the analysis has been investigated and will be discussed in this presentation.