4DVAR minutes 05/03/06
Hans, Dale, Qingnong, John M, Tom, John B, Chris
2:00-3:30 pm, 05/03/06, Lilly
Contents |
Review the progress in 4dvar made in the past two months
Real case runs
Hans spent most of his time on real case runs
After some initial tests on the AFWA 80x80x17 domain, he selected a 82x82x17 domain for the Katrina case. He also prepared a 91x73x17 domain for the Haitang (typhoon) case. New BEs have been computed over the two domains.
He also investigated the possibility of running long jobs on bluesky. Due to the wall clock limit and CPU charging rule, he decided at the end to make the long runs on G5s. On G5s, the current 4D-Var prototype runs as fast as on bluesky. However, a 4D-Var run takes 5days wall clock time to run 100 iterations.
The initial results did not look good. Two problems have been identified: errors in the Mactor projection and the difference between the ADP data and GTS data. He included the bugfixes for Macator projection into the 4D-Var branch. He also started experiments using GTS data.
Vertical diffusion
Qingnong started working on the vertical diffusion TL and AD. He had discussed the current code with Hans and Jimy. Hans and Jimy discussed the ECMWF “Buizza scheme”, which was in the first ECMWF 4D-Var. Some work seems necessary to improve the vertical diffusion before we move to a real PBL subroutine.
Merge 4D-Var branch
John B started merging the 4D-Var branch to the main WRF-Var trunk. The merging was delayed and slowed down due the recent KMA problems and AFWA MPP problems. His current estimate is to finish the merging work by the mid May.
Parallel the TL and AD
John M got the 4D-Var code and the testing case. He started the MPP work.
Discuss the 4dvar task table
- Hans went through the task table. He first pointed out a 4-month delay and hoped we can still meet the final deadlines.
- The participation of John M (0.5FTE) and Tom (0.25FTE) in the MPP and coupler tasks was confirmed.
- The merging task deadline was moved to 15 May.
- Hans discussed a little on the requirements on the nup and nup_inv. In the multi-incremental 4D-Var as we planned to build, the high resolution nonlinear trajectories need to be interpolated to low resolution using nup, and the low resolution analysis increments need to be interpolated to high resolution using nup_inv. John M wanted to have a clear description and Hans agreed to prepare a note.
- The coupler task was delayed. John M moved it after the MPP task. It can still be done before 8/31/06.
- John M started the MPP task on time. Dale agreed to provide the missing 0.25 FTE from the NPOESS project.
- Hans described the need, especially for 4D-Var, to remove C2A. Dale could do it but could not find time (~ one month). He also planned a code cleaning task, SE 0.5FTE, in NPOESS.
- The strategy for using TAF was discussed. Qingnong may spend 0.25FTE of his time on using TAF. The goal is to generate TL and AD for all remaining physics. Dale thought the expensive TAF license was not used efficiently and did not want to have it for the next year. John M wanted to look into the TAF usage issue first and make further decision later. Dale then suggested bringing in an experienced visitor in summer. He also mentioned positive experiences with TAF at other centers.
Below is the revised task table:
Need to replace Start, Mid and End, by the actual dates.
HH – Hans Huang JB – John Bray JM – John Michalakes QX – Qingnong Xiao DB – Dale Barker VS – Visiting Scientist (to be decided) PS – Project Scientist (to be decided) SE – Software Engineer (to be decided)
| Task | Type | Project | Duration | Notes |
| Merge 4D-Var branch to the main trunk | JB 0.1 | AFWA06 | Start - 2/15/06 | Delayed but started, expect to finish 5/15/06 |
| Vertical diffusion | QX 0.15 HH 0.1 | AFWA06 | Start - End | Ongoing. |
| Nup and nup_inv | JM+HH | AFWA06 | ||
| Coupler | JM 0.25 HH 0.1 | AFWA06 | 2/15/06-4/30/06 | Delayed. New deadline 8/31/06 |
| Parallel TL/AD | JM 0.25 TH 0.25 QX 0.1 | AFWA06 NPOESS06 | 5/1/06-8/31/06 | Ongoing. |
| Meteorological testing | HH 0.45 | AFWA06 | Start - End | Ongoing. |
| Deliver the prototype | HH+JB | AFWA06 | Done | |
| Report | HH | AFWA06 | End | |
| ARW-VAR | DB 0.25 | NPOESS | ?? | Start with removal of C2A |
| Generate TL and AD of necessary phys (Microphysics, Cumulus and PBL) using TAF | QX 0.25 VS | NPOESS? | ??-12/31/06 | Test of the generated codes will be started next year |
| JcDFI | VS 0.5 HH 0.1 SE 0.1 | NPOESS06 | Start - End | |
| Jbdy (1) | VS 0.5 HH 0.1 | NPOESS06 | Mid - End | |
| Jbdy (2) | VS 0.25 HH 0.1 SE 0.1 | NPOESS07 | Start - Mid | |
| More physics (1) 1. PBL 2. Micrphysics 3. Convection | VS 0.5 PS 0.5 SE 0.1 | NPOESS07 | Start - End | Need at least one more year (08) to complete. |
| Code optimization | SE 0.25 | AFWA07 | Start - End | |
| Long parallel runs | SE 0.25 HH 0.75 VS 0.5 | AFWA07 | Start - End | |
| More physics (2): 4. Surface Tuning existing phys; 5. Radiation; 6 GW drag | AWFA08 NPEOSS08 |
AOB
Qingnong and Hans should have a plan to write the sensitivity paper and the 4D-Var paper soon, as there are already requests on the 4D-Var code.
Hans mentioned a possible collaboration with the WRF chem group. There will be a meeting discussing this during the WRF workshop.
Dale mentioned Laura’s progress.
John Bray wanted a 4D-Var testing case, including the new BE for the domain. Hans agreed to prepare it.
Hans agreed to write the meeting memo and put it to the WRF-Var wiki page.