We envision a WRF "Lite" modeling system that uses the "Science
Module" code from the main WRF repository, but that is embedded in a
much more friendly framework and build system. The framework would take
advantage of the extensive higher level facilities found in the
Models-3 I/O API (M3IO), and would have much friendlier and simpler
driver and solver codes. At the same time, it should have greater
efficiency both for the parallel communication and for the
implementation of dynamics in the solver. We envision the development
as having two phases: first, hand coding the "Pilot Project" model
(complete with make system) implementing the NCEP model configuration
and using it to test various optimizations; then building a "Community"
model configuration system that constructs the driver and solver code
and make system (patterned after the code for the Pilot Project model)
for a model that satisfies a user's model-configuration specification.
We believe that in Phase 1, by combining the expertise of the staffs at
NCEP, BAMS, and NOAA/ASMD, we can construct a version of WRF that is at
the same time much friendlier and much more efficient than the current
NCAR versions. In the Phase 2 follow-on perhaps we would use additional
assistance in "code-writing" languages like python to construct a model
configuration and build system much friendlier and at the same time
more flexible and robust than NCAR's Registry/compile.
At the same time, we believe that both NCEP and BAMS have considerable
expertise at improving the performance of various critical science
modules. (The BAMS version of Reisner-1 microphysics is about five
times faster than its NCAR predecessor, for example, using optimization
ideas not to be found in any current WRF microphysics routine.) As an
independent effort, we suggest collaboration for developing selected
new science modules using our combined expertise.