The verification statistics from the data denial experiment have shown that the improvement in analysis accuracy is limited using the 2D-Var method, compared with TWN’s operational observation engine. In order to produce the surface analysis with further reduced error, a method using ensemble-variational hybrid data assimilation is proposed based on the GSI 2D-Var method in our York University/TWN research project. In contrast to the static background error assignment in TWN’s operational observation engine and the 2D-Var method, the hybrid method has the advantage of the dynamic and flow-dependent background error covariance because the model uncertainty from ensemble forecasts is considered. To assess the benefit that the dynamic background error covariance brings to the 2D surface analysis, a demonstrative system of the surface analysis using Ensemble Kalman Filter (EnKF) has been developed. As expected, the EnKF experiment results for the analysis variables of the 2-m temperature and the 10-m wind speed show the flow-dependent feature in the analysis increment field and the significantly increased accuracy in the analysis compared to the background.
The development based on the GSI system has been implemented to fulfill the hybrid method for the surface analysis within the CONUS domain at 3-km horizontal grid spacing. The system updates the surface analysis by assimilating METAR observations with the background field from the High Resolution Rapid Refresh (HRRR) model’s 1-hour surface-level forecasts. The model forecasts at the surface level of the Short Range Ensemble Forecast (SREF) from the National Centers for Environmental Prediction (NCEP) are used for the ensemble term’s input. The analysis variables include surface pressure, 2-m temperature, 2-m specific humidity, 2-m dewpoint temperature, 10-m wind speed, 10-m wind direction, and surface visibility. Some preliminary tests demonstrate that the background error covariance has the combined feature of both the static variational method and the dynamic ensemble method. Some ongoing improvement and optimization are being conducted. Once the hybrid system development is finalized, we will design and conduct the comparison experiment via running both the GSI 2D-Var system and the new GSI 2D-Var hybrid system for a certain period. NCEP’s operational product of Real-Time Mesoscale Analysis (RTMA) will also be compared with the results of our hybrid experiment.