Development and Testing of Hybrid Ensemble-Variational Data Assimilation for Cloud Hydrometeors in GSI

Successful data assimilation is of paramount importance for initializing numerical models, and Gridpoint Statistical Interpolation (GSI) is the data assimilation system used by the National Oceanic and Atmospheric Administration (NOAA) to assimilate observations in operational models. NOAA Earth Systems Research Laboratory (ESRL) Global Systems Division (GSD) developed an option within GSI for a non-variational cloud and precipitation hydrometeor (HM) analysis (hereafter referred to as the HM-analysis) to incorporate cloud/precipitation hydrometeor observations. This unique data assimilation approach improves analyses and forecasts of clouds and precipitation. In order to further improve prediction of clouds and precipitation, a transition of the HM-analysis to a variational and then hybrid ensemble-variational framework within GSI is now underway. This transition will also enable the HM-analysis to be used in any modeling system that interfaces with GSI, because no individual model drivers or libraries are required.

The new variational HM-analysis is applied initially to the hourly updating 13-km Rapid Refresh (RAP) and 3-km convective-allowing High-Resolution Rapid Refresh (HRRR). Single analysis case study experiments examining the impact of cloud assimilation will be presented. The cloud observations can be used to update cloud state variables directly and/or to update water vapor mixing ratio. The large sensitivity to observation errors and the balance between cloud and other observations types will be discussed. In addition, cycled retrospective experiments with the RAP will be shown. Results indicate a moist bias is present when cloud observations are assimilated, however qualitative comparisons with satellite observations show an improved representation of cloud fields. The assessment focuses on low-level cloud increments via ceiling analysis verification, moisture bias compared to aircraft and sounding observations, and cloud retention in the forecasts.