Friday, 21 July 2023: 9:30 AM
Madison Ballroom A (Monona Terrace)
The regional data assimilation team at EMC have been committed to the development
of the Gridpoint Statistical Interpolation (GSI) system for the first version of the Rapid
Refresh Forecast System (RRFS), with the EMC GSI authoritative repository on Github
being used by both collaborators and users at large of the Unified Forecast System in
this endeavor. With the RRFS implementation ambitiously targeting a large North
American domain at 3-km resolution, the optimization/refactoring of the GSI FV3-
LAM(Limited Area Model) interface for RRFS to improve on computational performance
has been a major focus. Recent improvements include the addition of the MPI
parallelization capability to the IO of GSI (including the conversion between fields on the
model and analysis grids) for model initialization/restart files, as well as OpenMP-based
enhancements. The scalability issues of the GSI using recursive filters (RFs) are also
being revisited in conjunction with the development of the highly scalable Multigrid Beta
Filter intended to replace RFs for background covariance modeling. Recent findings
on scalability and further optimization plans for RRFS GSI will be discussed.
of the Gridpoint Statistical Interpolation (GSI) system for the first version of the Rapid
Refresh Forecast System (RRFS), with the EMC GSI authoritative repository on Github
being used by both collaborators and users at large of the Unified Forecast System in
this endeavor. With the RRFS implementation ambitiously targeting a large North
American domain at 3-km resolution, the optimization/refactoring of the GSI FV3-
LAM(Limited Area Model) interface for RRFS to improve on computational performance
has been a major focus. Recent improvements include the addition of the MPI
parallelization capability to the IO of GSI (including the conversion between fields on the
model and analysis grids) for model initialization/restart files, as well as OpenMP-based
enhancements. The scalability issues of the GSI using recursive filters (RFs) are also
being revisited in conjunction with the development of the highly scalable Multigrid Beta
Filter intended to replace RFs for background covariance modeling. Recent findings
on scalability and further optimization plans for RRFS GSI will be discussed.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner