Testing and Evaluation of GSI Background Error Sensitivity for Regional Applications

The Air Force Weather Agency (AFWA) implemented the Gridpoint Statistical Interpolation (GSI) 3D-Var system into operations for their hemispheric domains in summer 2013 and will continue to implement GSI on their smaller regional domains through 2013. Initial verification statistics indicate good scores in the overall performance of GSI on the hemispheric domain. However, the smaller regional theaters show poorer forecast performance. In order to facilitate the improvement of forecast skill scores in the regional theaters, the Developmental Testbed Center (DTC) is conducting extensive testing and evaluation using a functionally similar environment to that used in AFWA operations. The end-to-end testing environment includes the community GSI v3.1 coupled with WRF-ARW v3.4.

Given the GSI implemented at AFWA is a 3D-Var system, specifying domain specific Background Error Covariance (BE) matrices is vitally important in improving forecast performance. The DTC has established two parallel real-time systems as well as a retrospective testing environment to test forecast performance sensitivity to different BEs. The newly developed beta version of the community GEN-BE 2.0 tool developed at NCAR/Mesoscale and Microscale Meteorology (MMM) is used for domain specific BE generation. Domain specific BEs are generated using the NMC method from several months of forecasts with additional tuning. This presentation will diagnose the various BE factors and discuss aspects of generating and tuning domain specific BEs using GEN-BE 2.0. Additionally, the forecast performance when using the domain specific BE relative to using pre-computed operational BEs will be presented.