Reports from Hans, Tom, Xin, Yongsheng, Jimy and Qingnong were compiled as the "meeting memo" for May 2007.

Progress review

Optimization

TH continued working on optimization of 4DVAR adjoint via trade-off of re-computation for memory. TH completed two more rounds of trade-offs and removed unneeded computation from two routines. For the serial CWB 135km case on bluevista (IBM), run time for solve_em_ad() has been reduced by more than 25%. Note that this measurement does not include other 4DVAR components such as TL and NL.

TH discovered that 1-task and 16-task parallel (MPI) runs of solve_em_ad() did not produce bitwise-identical results as they should. Instrumented solve_em_ad() to detect parallelization errors. Tracked down and fixed several parallel bugs (missing halo updates, incorrect loop indices when computation in the halo region is needed, etc.). Corrected code now can produce bitwise-identical results from solve_em_ad() for the 135km CWB test case.

TH began testing these optimizations with a parallel build with help from XZ. They correctly set up a parallel configuration on blueice. TH added his optimizations to the corrected parallel code, built, tested, and verified that the optimizations do not affect results (i.e. adjoint output is bitwise identical with and without my optimizations).

XZ discovered that the new WPS-generated input files expose a previously undetected bug in solve_em_ad(). TH is now investigating this error.

XZ and HH made much progress in the multi-incremental formulation. Individual subroutine coding is finished and now they are searching bugs and assembling codes and scripts together.

4D-Var Met test

XZ re-generated the typhoon Haitang case using WPS. HH and YC re-visited the Haitang case using the latest WRF 4D-Var configuration and the WPS data. They met problems with the 4D-Var scores. Work is still ongoing to understand whether the latest WRF 4D-Var configuration has problems or the WPS data does.

YC, XZ and HH recovered the single observation experiment with 200605 wrf4dvar version and verified it with the wrf4dvar 2.2 version.

YC and HH made experiments with both parallel and serial versions of 4D-Var using Typhoon Haitang case. The evaluation is on going.

YC prepared a new case, Typhoon Shanshan, to compare the 4D-Var with 3D-Var (and EnKF, run by Hui Liu).

Develop and test the control of lateral boundary condition (Jbdy) perturbations in WRF-Var.

YC investigated some lateral boundary related problem, using the single ob environment. There are clearly noises along the lateral boundaries. The lateral boundary control formulation will start in November.

Include additional physics in the WRF linear and adjoint models, and test in 4D-Var.

JD has almost completed the work on a simple cumulus scheme. The scheme itself is a minimal code of about 100 lines designed for easy linearization and adjoint development. Current work is on finalizing a few parameters of the scheme mostly related to its updraft mass flux and internal rain evaporation rate, and testing the sensitivity to them.

Results are promising in tests of the standard small-domain June 2001 convective test case at 10 km grid size, and comparison with the other available schemes in WRF (KF, BMJ and GD) shows that it falls within the range of variability of these schemes.

QX will soon start working on developing TL and AD for the simple cumulus scheme.

AOB

The next meeting: 10 am 6/29/07, room 2072.