segmentation error only when running high resolution domains (333m)

[juanan]

Hi,

I have the following oonfiguration for my simuation .
I am using ERA5 data
The configuration has 4 domains. If I run the 4 domains it will provide a sementation error, while if I only run the 3 lowest resolution domains (not the 333 m), the model with run smoothly
Why can't the mode dela with the high resolution domian? Any suggestions?

thanks in advance

 

[Ming Chen]

How long did the case run before it crashed with seg fault?
Did you see any CFL violation or other error messages when running over 4 domains?
Where are your model domains located, especially the D04 domain?

 

[juanan]

thanks for the response.
I attach the rsl.error file. The crash happens at the same time independet of the number of cores, after reachin 9 min of real time. However, the CFL violation appear in the first time step, and as said before only when including domain 4 (333 m resolution).
I attach the rsl.error file when including max-dom=4 of the previous namelist.input.
The 4 domains area centrered in Singapore (tropics). Domain 4 has some mountains (not very high) in the border of the domain... These mountains did not affect simulation in version 3.8, although another physics where used

 

[Ming Chen]

CFL violation indicates the model integration is unstable. There are a few issues here:
(1) you run with "tropical" suite. However, for grid intervals smaller than 3km, cumulus scheme should be turned off
(2) the resolution of 333.33m is within typical grey zone resolution for PBL scheme. PBL assumptions are no longer valid, which may lead to model crash
(3) specify dx = 9000, 3000, 1000, 333.33, and the same for dy
(4) turn on w_damping, and set diff_opt = 2, 2, 2, 2
Please change above options in your namelist and try again.

 

[juanan]

Hi:

I did all the changes (see the attached files) and the result/errors (forrtl: severe (174): SIGSEGV, segmentation fault occurred)
are the same in the same point in time. The cumulus physics was on for the 9km grid
Any other ideas? thanks a lot for the feedback

 

[Ming Chen]

Do you still see CFL violation?
If not, please make sure you have sufficient memory to run this case. Seg fault may also be induced by memory issue.

If there is no CFL violation and you have sufficient memory to run this case, then the model crash is possibly due to physics. As I mentioned earlier, the resolution of 333.33 meter could be an issue for PBL schemes.

To figure out excactly what is wrong, please recompile WRF in debug mode, i.e., you need to issue the commands ./clean -a and ./configure -D, then recompile all the codes.

By this debug mode, you can find when and where the model first goes wrong, which will give you some hints how to fix the issue.