The HRRRDAS system utilizes regional domains, which benefit from the periodic input of global scale information to account for large scale weather patterns. To this end, the HRRRDAS relies on the Global Forecasting System (GFS) for mean initial conditions. The frequency in which GFS data are used is an important question. The initial HRRRDAS configurations restart once per day with GFS initial conditions. This configuration is technologically convenient, but can have the weakness of a discrete change in ensemble characteristics with the input of the new GFS data. This deficiency has a clear impact on the downstream deterministic HRRR forecasts (e.g. degraded reflectivity skill). Recently, experiments with a “rolling ensemble” approach have been explored. The rolling ensemble concept uses GFS data 4 times per data for ¼ of the ensemble members. This configuration removes the discrete change, provides global information every 6 hours, and can cause the ensemble to be less gaussian. Experiments with both configurations will be presented and discussed.
The HRRRDAS system uses the GFS Data Assimilation System (GDAS) to provide initial ensemble perturbations. Similar to the mean initial conditions, experiments in which large scale perturbations periodically replace the storm-scale perturbations in some or all members are conducted. This approach can be compared to maintaining storm-scale perturbations and “re-centering” the ensemble. Additionally, spread maintenance through inflation to prior spread is included in the HRRRDAS. This presentation will share recent ensemble spread results for cycled data assimilation and for the use of ensemble covariances in the 3DEnVAR HRRRv4 data assimilation. Cases and statistics for severe weather events in Spring 2019 will be presented.