Thompson Scheme (posted 12/13/10)
Problem: WRF Versions 3.2 and 3.2.1 contain a MAJOR bug in the Thompson et al
microphysics option number 8. The bug applies for ALL simulations using this option but is worst for high resolution convective cases because it involves the production of graupel from rain and snow collisions. Any users of this scheme should not trust the results of simulations using these code versions because, instead of creating graupel due to rain/snow collisions, the bug was causing the removal of graupel. This bug is NOT found in earlier versions of WRF.
Besides the bug fix for graupel, other changes have been made to the scheme including a more subtle bug related to rain evaporation that has existed in ALL WRF code versions containing the Thompson et al (2008) scheme since late 2006. This bug was minor in comparison but is fixed in the new code. Any users wishing to fix the rain evaporation bug (involves 2 lines of code) in previous WRF versions may contact the author for details to include this change only.
Also, the intended bug fix replacement file contains a new method to determine the graupel Y-intercept parameter as a design change as well as changes to handle mass/number imbalance checks, especially related to sedimentation. The latter changes are better mathematically and should be hardly visible in real simulations, but came about due to testing the sedimentation scheme in isolation from the rest of the code. The bug-fix code should also be released as part of the next standard WRF code release, though further changes are possible between now and Spring 2011.
Solution: Download a new copy of module_mp_thompson.F.fixed and replace the one in phys/ directory. Recompile.
Uninitialized Variable Problem (posted 5/5/10)
Problem: A problem with potentially uninitialized ProcOrient field in linked lists of state variables is uncovered in V3.2. This could lead to sporatic hanging of MPI jobs.
Solution: Download gen_allocs.c file and replace the same file in WRFV3/tools/ directory, and do a full recompile that begins with 'clean -a'.
Grid-nudging File Generation Error (posted 4/20/10, updated 9/7/10)
Problem: There is an error introduced in V3.2 that causes some fields being incorrectly zeroed out in the real program when writing out the wrffdda files.
Solution: download reg_parse.c file and replace the one in WRFV3/tools/ directory. Do 'clean -a' (note, one must use '-a'), and recompile.
If you are compiling with WRF-Chem, there is another change needed in Registry.EM_CHEM: search for string 'igr', and change it to 'i{10}r'. Do 'clean -a', and then recompile.
Additional namelist Required To Specify IO Format Option (posted 4/20/10)
Due to expanded IO capability in V3.2 (runtime IO and increased IO streams), several runtime options require additional namelist.
1. sst_update: it requires io_form_auxinput4 = 2 in &time_control
2. find_input_stream: it requires
io_form_auxinput2 = 2 in &time_control
Other namelists like io_form_history, io_form_input, io_form_boundary and io_form_restart are also required in the namelist.input.
Possible Performance Hit with V3.2 code (Posted 4/2/10)
1: Two fixes introduced in V3.2 may result in performance hit. One of them is the change for RRTM longwave radiation option. Since the fix involves adding model layers above model top, the lower the model top (p_top), the more likely to see a slow down.
Solution: One solution is to increase the time interval to call the scheme hence maintain the same wallclock time. The other solution is to modify phys/module_ra_rrtm.F, line 31, and change layer depth deltap from 4 mb to a larger value. However with larger layer depth, one may not obtain optimal results.
2: The other performance hit is related to adding semi-lagrangian fall terms in all WSM/WDM schemes. This could affect finer grid model run more.
Compile V3.2 with ifort (posted 4/2/10)
With the new code, one must ust at least ifort 10 to compile
Restart Issue (posted 4/2/10, updated 8/18/10))
Problem: There are problems with restart produing idential results as continous run for physics options level 3 MYNN PBL, and BEP urban option.
Solution: Do not use restart when these options are used.
OpenMP And MPI For Some Physics in V3.2 (posted 4/2/10)
Problem: Certain physics options do not work correctly with OpenMP or MPI compile. These are
- mp_physics = 9 (Milbrandt-Yau scheme) (OpenMP)
- sf_urban_physics = 2 (BEM) (OpenMP)
- ishallow = 1 (MPI)
There is no known problem with these physics options if OpenMP or MPI is not used.
Solution: Do not attempt to use the above listed options with OpenMP or MPI compile.
