4DVAR minutes 10/31/06
10:00 am 10/31/2006
Lilly room
Participants: Dale, Hans, John B, John M, Michael, Qingnong, Tom, Xiaoyan
Contents |
Progress review
Merge 4D-Var branch to the main trunk
Xin, with the help of Michael and John B, got the first version of the "3 exes" system working with the wrftrunk, wrfvartrunk and wrfplus. The system was only tested with serial version and on bluesky.
Xin still needs to get JcDFI into this system.
Xin needs John M's help to get the 3 exe system run in parallel mode.
Veritical diffusion and large scale condensation
Jimy made a simplified PBL scheme. Qingnong and Xiaoyan helped Jimy test the code and compare it with other PBL schemes.
Qingnong made the TL and AD of S-PBL using TAF.
The TL is under testing. The AD test will follow. The testing results will be reported next time. The final test should be done within the 4DVAR framework.
Qingnong discussed with Jimy on a simplified large scale condensation scheme. Jimy told Hans that he would make such a scheme soon. It is agreed at the meeting that Jimy should be invited to the future 4DVAR meetings.
We had a long discussion on how this work should continue.
- Hans: talk to Jimy to get his code into the wrf trunk, so that Tom can include it in the simplified nonlinear model (SN) and generate TL and AD using TAF. Tom needs to create a branch from the current freezed trunk.
- Qingnong and Xiaoyan: before Tom's TL/AD, still implement Jimy's schemes in wrfplus, generate and test TL/AD in his usual way, and conduct 4DVAR tests (Hans can help in this).
- Qingnong and Tom: converge the TL/AD generation and testing.
- Hans planned to have the S-PBL in the AFWA release. John M expressed his worries about parallelization of wrfplus after S-PBL implementation. Hans thought most of the phys routines should not affect parallelization and agreed to talk to Jimy about this.
TAF
Tom made further progress in the source to source software. The software can create a SN like Qingnong's manually developed SN. However there are still problems when using TAF on the whole SN. Tom got more timely response from FastOpt in the past weeks.
Tom cannot make a trip to Germany in December, but will contact FastOpt to find out whether he can go there in January. However, the TAF license will be expired at the end of December. Tom will find out whether there is any problem for him to visit FastOpt when we no longer hold the license.
Nup and nup_inv
Xin got a small program from Wei for simple interpolation. He tested it and belives he can develop nup and nup_inv based on this program. John M agreed to review Xin's code.
This should be included in the multi-incremental formulation task next year.
Meteorological testing
The Haitang case (4 cycles per day for 3 days) has been re-run with the centered assimilation window. Comparison is ongoing and should be reported next time.
In the KMA project, we have a task to compare 3DVAR and 4DVAR. A KMA case was selected and initial tests started.
JcDFI
Xin got the CG problem solved. The next step is to get JcDFI work in the 3 exe system.
General discussion about 4DVAR cost
The current timing for an "operational like" 4DVAR setup is about 5 hours. Hans wants to have 10X reduction of it to 0.5h so we can afford to have more vertical levels (2X) and some physics. John M guesses the best we can hope is a 2X (not 10X) reduction. With the grid chosen, even the nonlinear model does not scale well. Dale wants to have a clean comparison between NL (without phys) and AD. John B wants to see the trunk wrf timing within the 3 exe system.
Hans will discuss with Jimy on other possible simplifications in dynamics, if there is any.
AOB
Michael and Jimy are added to the 4DVAR developer list.
There is no comments on the last meeting memo.
The next meeting: 10:00am, 11/30/2006.