4DVAR minutes 022807
10:00a, room 2072
Participants: John M., Tom, Hans, Yongrun, Jimy, Qingnong, Xiaoyan
WRF-Var (Hans)
To avoid further confusion to the outside users, we need to find a better word than WRF4DVAR when we refer to the 4-dimensional component of WRF-Var. Within the group, we can just use 4D-Var before we find a better name. Should not use WRF4DVAR and WRF3DVAR in paper titles and web page headings.
Hans will ask John Bray to change the 4D-Var page title.
AFWA and NPOESS 2006 final report (Hans)
Hans checked with Dale and found out that the report is not an AFWA project deliverable.
John M made a ppt for the parallel and coupler tasks.
Qingnong and Xiaoyan made a ppt for VDIFF and LSC tasks.
Xin and Hans are writing a report for JcDFI and meteorological tests tasks. Almost done. Will distribute to the group.
Progress review
TAF, OpenAD and related (Tom)
Discovered that the automatically-generated tests for tangent-linear (TL) and adjoing (AD) codes include routines not tested via the hand-generated tests in module_check.F. Unfortunately, some these tests fail. I asked Qingnong Xiao and Thomas Nehrkorn if this might be due to the global summation used to compute the test results (i.e. sums over all elements of all arrays tested) hiding failures because of limited precision (even with 64-bit reals). They both independently confirmed that this is possible. Each of these failures must be investigated. Thomas Nehrkorn sent examples of his testing methods. Created branch https://svn-wrf-model.cgd.ucar.edu/branches/4DVAR_TAF starting from revision 1874 of https://svn-wrf-model.cgd.ucar.edu/trunk to hold this work in the svn repository.
Discussed OpenAD with Jean Utke and Paul Hovland. Jean Utke agreed to run OpenAd on "auto-simplified" WRF code. I made further modifications to hide Fortran derived data types (these will be automated later if needed) and sent sample code to Jean for testing. Jean has run some initial tests which have exposed some bugs in OpenAD, which are now being fixed. Also discussed OpenAD with Chris Hill (MIT) who is using both TAF and OpenAD for the MITgcm.
Xiaoyang Zhang found a bug in my auto-generated tests for subroutine lscond(). I fixed the problem. This bug is also present in some of the hand-coded tests in module_check.F, including those that I used as a template when building the automatic test generator. I have re-run the fixed automatic test generator and will provide fixes for module_check.F to Qingnong Xiao and Xiaoyang Zhang. In addition, I have found another possible bug and will investigate with Qingnong and Xiaoyang.
Physics (Qingnong)
Finally, the TL and AD of LSC passed all the test. The bugs were due to the inconsistency between different versions. Will use the trunk code for the next task. The code will be sent to Xin later today.
Meteorological testing (Xin and Hans)
Finished the combination of WRF 4DVAR related codes with trunk codes of WRF and WRFVAR. Almost all modified codes are lodged.
Finished the coding and testing of single processor version on Mac and parallel version on bluesky, bluevista and blueice.
Finished the experiments of real cases with the WRF 4DVAR system and work of the final report of subcomtract.
AFWA 2007 4D-Var work plan
| Priority | high | medium | high | high | low | low | |
| Multi-inc | J_lbc | Optimize | Met. test | convection | A2C | Sum | |
| Barker | 0.15 | 0.15 | |||||
| Huang | 0.1 | 0.1 | 0.1 | 0.5 | 0.80 | ||
| PS-I | 0.2 | 0.20 | |||||
| Xiao | 0.3 | 0.30 | |||||
| Dudhia | 0.1 | 0.10 | |||||
| Michalakes | 0.1 | 0.10 | |||||
| Henderson | 0.4 | 0.40 | |||||
| Zhang | 0.25 | 0.25 | 0.25 | 0.75 | |||
| Sum | 0.35 | 0.35 | 0.85 | 0.7 | 0.4 | 0.15 | 2.80 |
Multi-incremental formulation (Zhang 0.25 FTE + Huang 0.1 FTE)
Two new programs (may combined to one executable) will be coded first: 1) nup - interpolate from fine resolution to coarse resolution; 2) nup_inv - interpolate from coarse resolution to fine resolution.
Extend WRF 4D-Var version 2 system with the multi-incremental capability using nup and nup_inv.
Period: 1 March – 31 May 2007
Optimization (Henderson 0.4 FTE + Zhang 0.25 FTE + Michalakes 0.1 FTE + Huang 0.1 FTE)
General support for WRF 4D-Var parallelization (Michalakes).
Period: 1 March 2007 – 31 January 2008
Make an option of wrfplus using more memory to reduce re-computations. The first working version will be ready for testing before May. The final version will be ready for testing before September. The goal is to reduce the current 4dvar wall clock time by 50% (Henderson).
Period: 1 March – 31 August 2007
Reduce nonlinear re-computations of the basic states in the tangent linear and adjoint integrations. The goal is to reduce the current 4D-Var wall clock time by 50% (Zhang and Huang).
Period: 1 June – 31 August 2007
Meteorological tests (Huang 0.5 FTE + PS-I 0.2 FTE)
Make a 45 km 4D-Var configuration on the AFWA 15 km domain. Collect observations for one or two selected cases (a case here should cover at least a few days). Make the configuration as close as possible to the AFWA current operational system. Compare the 4D-Var results with 3D-Var and make necessary tuning.
Make a 15/45 km 4D-Var configuration over the same domain and cases. Run the multi-incremental (15-km outer loop and 45-km inner loop) formulation and compare the results with the 15 km 3D-Var.
Provide a pre-operational setup to DATC for pre-operational tests. Support and interact with DATC colleagues during the pre-operational tests.
Experiment with different domain sizes and different resolutions to provide guidelines for possible changes of the pre-operational configuration.
Period: 1 March 2007 – 31 January 2008
Lateral boundary control (Zhang 0.25 FTE + Huang 0.1 FTE)
Study different formulations for lateral boundary control used by limited area data assimilation systems. Code the selected formulation and start the initial tests. Testing may continue to next year.
Period: 1 March 2007 – 31 January 2008
Convection (Xiao 0.3 FTE, Dudhia 0.1 FTE)
Investigate the current available convection schemes and select the most efficient one. If necessary, implement it in WRF and compare it with the commonly used scheme by WRF users.
Develop tangent linear and adjoint code of the selected convection scheme.Implement it in wrfplus.
Further testing will be carried out if an additional funding of 0.1 can be made available.
Period: 1 March – 31 August 2007
Using C-grid in Var (Barker 0.15 FTE)
Develop an option for WRF-Var to use C-grid and assess the impact on computational and meteorological aspects.
Period: 1 March – 31 August 2007
4D-Var plan related issues
Tom will contact Thomas Nehrkorn to discuss the sharing of the 0.4 FTE task on optimization.
The AFWA 2007 4D-Var work plan is the major part of our 4D-Var work plan. 4D-Var work from other projects should also be included and coordinated. Qingnong will provide his plan for the 4D-Var work in a hurricane project. Hans will ask Jenny to provide her plan for the 4D-Var work in the CRIEPI project.
AOB
The end of March is in the spring break. We will not have a project meeting. However, everyone involved in 4D-Var should sent a short progress report to Hans before March 31. Hans will put all the reports together.