Since Spring 2015, an experimental, real-time, full-CONUS, convection-allowing (3-km horizontal grid spacing) ensemble forecast system has been running at NCAR. Daily 10-member ensemble forecasts are initialized at 0000 UTC and integrated through 48-hours. This system is unique in that the forecasts are initialized from a continuously cycled, limited-area, 50-member, mesoscale ensemble Kalman filter data assimilation (DA) system using the Data Assimilation Research Testbed (DART) toolkit.

This presentation will provide a brief description of the system design and discuss preliminary performance of the DA system and forecasts, with a focus on the seasonal aspects of performance. Efforts to enhance both the skill and reliability of the ensemble forecasts through improvements in the cycled EnKF DA system will be described. These efforts include an ongoing investigation of various algorithms for spread maintenance in the DA system, such as adaptive prior inflation, posterior inflation, combined prior and posterior inflation, spread restoration, and relaxation to prior spread. A brief description of the posterior inflation and spread restoration will be included. We will also discuss results regarding analysis and forecast sensitivity to the moisture observation type and observation error treatment, number of model levels, and boundary layer physics scheme. Finally, future plans to improve the forecasts and DA system will also be discussed.

---

*email: sobash@ucar.edu
*Preference: Oral