rdlai2d=true conflicts with NoahLSM

[Xun_Wang]

Hi,

I am using the newest WRF v4.0.1 and found the following problem:

The LSM I am using is Noah LSM and the sst_update is turned on. I want to use LAI from input instead of from table. When rdlai2d = .true., the model crashed after a few second and I got the following massage in the rsl.error file (I already set 'ulimit -s unlimited' before running wrf.exe):

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image PC Routine Line Source
wrf.exe 00000000030E2A1D Unknown Unknown Unknown
libpthread-2.24.s 00002B04EAE820C0 Unknown Unknown Unknown
wrf.exe 0000000001FDD2C8 Unknown Unknown Unknown
wrf.exe 0000000001FD6FCE Unknown Unknown Unknown
wrf.exe 0000000001FD1F92 Unknown Unknown Unknown
wrf.exe 0000000001FD0762 Unknown Unknown Unknown
wrf.exe 0000000001948D3E Unknown Unknown Unknown
wrf.exe 0000000001A3996F Unknown Unknown Unknown
wrf.exe 00000000013E3C29 Unknown Unknown Unknown
wrf.exe 0000000001262A40 Unknown Unknown Unknown
wrf.exe 000000000054C0D3 Unknown Unknown Unknown
wrf.exe 0000000000405601 Unknown Unknown Unknown
wrf.exe 00000000004055BF Unknown Unknown Unknown
wrf.exe 000000000040555E Unknown Unknown Unknown
libc-2.24.so 00002B04EB3B22E1 __libc_start_main Unknown Unknown
wrf.exe 000000000040542A Unknown Unknown Unknown

However, when I turned rdlai2d off, the model could run successfully. Also, when I changed the LSM to Noah_MP (dveg=9, which also mean use LAI from input), there was no segmentation fault. I also had this problem when using WRF v4.0.

Cheers,
Xun Wang

 

[Ming Chen]

Would you please upload your namelist.wps and nameless.input for me to take a look? Thanks.

Ming Chen

 

[Xun_Wang]

Dear Ming,

Thanks for your reply! Here are my namelists.

Cheers,
Xun

 

[Xun_Wang]

Some updates,

1. Before the model crashed, there are warnings like "WARNING, negative qv -8.7665268E-07 at i,j,k= 56 99 3" . But this kind of warning also exist when rdlai2d=false and didn't cause any problem.

2. I reduced the history_interval to have output every two time steps, then I found some strange grid points in surface field such as T2, Q2, HFX, LH. But the locations of these points are not the same as the locations where these is negative qv.

3. I used mpirun with 28 processors and it worked well except when rdlai2d=true. Then I just ran ./wrf.exe and it worked. The mpi pacakge I use is mpich-3.0.4

 

[Ming Chen]

I don't think a very small value of negative qv is the reason for the case to crash.
I repeated your case and the model crashed due to CFL violation. However, after I reduce the time step to a small value, the case still crashed. This makes me believe some physics might be wrong.
I will continue to look into this case, but it may take time. Thanks for your patience.

Ming Chen

 

[Ming Chen]

I attach a revised code module_sf_noahdrv.F. Please use this code to replace the original one in WRFV4.0.1, then recompile WRF.

Please let me know how it works. Thanks.

Ming Chen

 

[Xun_Wang]

I tried and it worked! Thanks for your help!

Xun Wang