4DVAR minutes 07/14/06

Participants: Dale, Hans, John B, John M, Qingnong, Tom, Xin

Contents

[edit]

1. Review and discussion on the task table

[edit]

Merge 4DVAR branch to the main trunk

John B got the merged code working for 3DVAR and resumed the 4DVAR work, but has got heavily bogged down with the number of changes. As a shortcut, Hans agreed to help John with the one-sided FGAT, which does not need the TAF-generated TL and AD, with that to be tested as part of the end September release. The TAF changes discussed later means that the existing TL and AD code need not be merged into the WRFVAR trunk.

[edit]

Vertical diffusion

Qingnong, Jimy and Hans had a meeting recently. The vertical diffusion in dynamics works, but a constant K is used. Qingnong has been working on a verically varying K(z). In the same time, Hans agreed to work with Jimy (next week) on the surface coupling.

[edit]

nup and nup_inv

No progress.

[edit]

Coupler

John M has got the coupler done! He stressed that this version still uses disk IO and relies on IBM LoadLeveler.

[edit]

Parallel TL/AD

The parallel code compiles and runs, but its produced results are still not quite right. John M has been debugging the code lately. Qingnong agreed to help John M.

[edit]

Meteorological testing

Hans introduced the cv5 option, with Yongrun's help, into the 4DVAR branch. He made a series of experiments to compare the meteorological performance of 3DVAR, FGAT and 4DVAR, all using cv3. He is repeating these experiments with cv5 and should finish them next week. The next step is to run these experiments with centered data assimilation window.

[edit]

JcDFI

Xin arrived two days ago and started with Hans on the work already. Xin will be here for 3 months.

[edit]

Physics TL/AD

We decided to discuss this issue in a seperate meeting.

[edit]

2. Strategy discussion

Throughout the task review several issues related to short-term and long-term strategy were discussed.

Should we buy the TAF license for the next year?

Should John M continue the current debugging work on the parallel code?

Should John B merge the current TL and AD, based on a 18-month old WRF version, or new TL and AD of the new WRF version? The latter requires TAF work.

It was decided for John M to finish the manual parellelisation of the 4DVAR branch, as a backup for the Feb AFWA release, but to concentrate effort on regenerating a new S/NL, TL and AD code from the latest WRF trunk, incoporating parallelism before TAF is run to regenerate the code. Taking advice from other centres that use TAF, and feeding back problems we should aim to produce code that does not need manual fixes, and define a coding and testing process that would allow future releases to be produced with minimal effort. This would allow us to assess whether to renew the TAF license, or maintain the code manually in future.

In discussion after the meeting a provisional tasklist was generated:

  • Add transpose routines to RSL_LITE, so RSL can be retired. John M end August
  • Add adjoint routines to RSL_LITE. John M end August
  • Complete manual parellisation of 4DVAR branch as backup. John M end August
  • Merge WRF and WRFVAR codebases, excluding TL and AD code. John B, end September
  • Extend WRFVAR test rig to handle 4DVAR cases. John B, end September
  • Produce new automatic TL AD models for AFWA release end Feb 2007.
    • Update S/NL model to WRF trunk. Tom, Qingnong
    • Add parallel mods to source code. John M
    • Run TAF to produce TL and AD models, debug and test until minimal manual fixes needed. Tom
    • Get testimonials on how other sites use TAF, from people like Nehrkorn and Da Silva. Hans
    • Coupler. John M
    • Coding of the above to be finished by Christmas, to allow Hans to do scientific testing before release

NB

  • In discussions after the meeting, it was agreed that it made sense of [[John B] to take over Tom's parallel global WRFVAR and radiance data load balancing, to allow Tom to concenrtate on TAF work. Mike Duda is available as a resource in the autumn to assist with the TAF work too.
  • AFWA need to be informed of this strategy at the end August mid-term review, as it may be in their interests to wait for a possibly delayed new 4DVAR after Feb rather than work with the more limited backup solution.
[edit]

3. AOB

Hans agreed to update the task table after the meeting.

The next 4DVAR meeting will be at 2 pm 9/14/2006.