Development and research of GSI-based Var/EnKF/hybrid data assimilation system for convective scale weather forecast: A comparison of GSI-based EnKF and 3DVar for multiscale analyses and forecasts

One unique challenge in convective scale forecast is that the accuracy of the forecasts depends not only on processes at the convective scale but also on the mesoscale and synoptic-scale environment supporting them. Therefore, accurate forecasts for convective scales require data assimilation systems to properly estimate states from multiple scales. A GSI-based data assimilation (DA) system including 3DVar, EnKF and hybrid are extended to the multi-scale assimilation of both conventional observations and radar radial wind and reflectivity observations. This study focuses on systematic comparison of GSI based 3DVar and EnKF in the context of multiscale data assimilation where scales ranging from convective scales to synoptic scales were resolved by both the model and the observations. Such comparison will facilitate understanding on how the differences of different DA techniques lead to differences of analysis at various scales and therefore impact subsequent storm scale forecasts. Diverse cases are used to obtain robust averaged results. The cases include many examples of both discrete cellular convection and organized Mesoscale Convective Systems (MCSs). Averaged over 10 cases, convective scale precipitation forecasts initialized by GSI-based EnKF were much more skillful than GSI-based 3DVar. The positive impact of assimilating radar data lasted for 5 hours when assimilated by the EnKF. In comparison the impact only lasted for 1 hour when assimilated by 3DVar. A case study was examined in details. It was found that better storm scale forecasts by EnKF were attributed to better analyses for both storm scale and mesoscale environment. The mesoscale analysis results in a better precipitation forecast for EnKF than for 3DVar due largely to a better analyzed warm front for EnKF. While the EnKF forecast with radar DA remains better than the forecast without radar DA throughout the forecast period, the 3DVar forecast quality is degraded by the radar DA after the first hour. The 3DVar analyzed storms are not maintained in the forecast and excessive cold pools emanating from the collapsing storms further degrade the forecast at later lead times. These features are attributed to the lack of cross-variable correlation in the static background error covariance for 3DVar.