A Regional Ensemble Kalman Filter Data Assimilation System Employing GSI Observation Processing and Initial Tests For Rapid Refresh Forecast Configurations

Abstract

A regional ensemble Kalman filter data assimilation system based on the latest version of Gridpoint Statistical Interpolation (GSI) and an Ensemble Square Root Filter (EnSRF) package configured for the NCEP Global Forecasting System has been established for possible future application in the hourly updated Rapid Refresh (RR). The planned operational Rapid Refresh system will use the Weather Research and Forecast (WRF) prediction model and the GSI 3DVAR analysis combined with a cloud analysis package and diabatic digital initialization (DDFI) procedure. The RR system will perform hourly data assimilation cyclones and uses a 13-km resolution North America grid. The current initial effort intends to compare the performance of the EnKF analyses produced at ~40 km resolution (1/3 the resolution of RR grid) with that of GSI analyses, using the same sets of observations. As indicated earlier, all observational preprocessing, quality control as well as the calculations of all observation innovations are performed by the GSI. The observational data include all those used in the RR system, including surface data, sounding, wind profiler, satellite retrieved winds and temperature and satellite radiance. Radar data are used in some experiments. The quality of analyses will be evaluated mainly based on resultant forecasts produced on the 13 km RR grid. The ~40 km EnKF ensemble mean analyses are interpolated to the 13 km RR grid for forecast initialization. Forty ensemble members are typically used. As for the standard RR system, hourly analysis cycles are performed, and a two-week-long period is chosen for continuous assimilation experiments. The first pair of experiments using EnKF and GSI DA systems, respectively, will exclude the use of radar data, because of the relatively coarse resolution of the EnKF analysis grid. The second pair will introduce radar reflectivity and radial velocity data on the 13 km RR grid, using the GSI 3DVAR and cloud analysis procedure, as an add-on step, in combination with the DDFI procedure within the WRF model. The forecasts resulting from the 4 sets of forecasts will be verified against observations, for both precipitation and routine meteorological variables. Furthermore, the relative performance of perturbing the lateral boundary conditions of the EnKF system using a global EnKF system or perturbations sampled from the WRF 3DVAR background error covariance will be examined.