Scheduled Downtime
On Friday 21 April 2023 @ 5pm MT, this website will be down for maintenance and expected to return online the morning of 24 April 2023 at the latest

WRF model stability check

mgolbazi

mgolbazi

New member
Hi everyone,

I am running a real WRF-LES case where I see many non-physical lines and waves in my wind speed (attached) and temperature plots which indicate that my run has gotten unstable at some point. This is not the first time that I am experiencing this (I have experienced the same in mesoscale runs) and I was wondering if there is a method (like a script or a computational method) to check and see if the run was stable?

In other words, how can I ensure that there weren't instabilities in my model run without visualization?

Thank you,
Maryam
 

Attachments

  • ws_les-d03-702_036.000002.png
    ws_les-d03-702_036.000002.png
    557.1 KB · Views: 4
Dear Ming,

I attached my namelist to this post. I would appreciate your comments on this configuration.

However, my question is regardless of the namelist setup. I am wondering if there is a parameter that tells us about model instability. For instance, I know that the CFL errors are an indicator of the model getting unstable. Does this mean that the lower the CFL the more stable the model is?

Thank you,
Maryam
 

Attachments

  • namelist.input
    8.1 KB · Views: 3
Last edited:
Maryam,

CFL violation often indicates that the model is numerically unstable at certian points. However, if wrf.exe can run to the end successfully, it implies that the model overcome the instability and we usually don't worry in this case.

I could see the small wave structures in your result, which might be attributed to 2*dx wave disturbance.

Would you please add diff_6th_opt = 2, 2, 2 to your namelist.input, then rerun this case?

Please let me know whether the result can be improved.
 
Ming,

Thanks for the reply. I had already set the diff_6th_opt to 2 in the namelist I sent you.

I also tried "gwd_opt =3" and reducing the time step to 9 but the problem persists. I can also see the streaks in the parent domain (d02) which has a resolution of 750m. My parent-to-nest ratio for the LES domain is 5 and I am wondering if that is causing the problem. I know that 750m resolution for d02 falls into "terra incognita" but so does 450m (a ratio of 3 instead of 5). I am not sure what is the best way to skip this.

Please let me know if you have any other suggestions.

Thank you,
Maryam
 
Maryam,

The only suggestion I have at present is to set diff_6th_factor = 1, which will completely remove 2*dx wave.

I am not sure whether this works for you. We don't have many experiences running WRF at such a high resolution like you did. We did see the strip structure before, and in some cases the diff_6th option can successfully remove them.

Please try and let me know how it works. Thanks.
 
Thanks! I tried setting diff_6th_factor to 1 but the model keeps crashing with CFL errors. I tried w_damping, decreasing the time step, and increasing the epssm to 0.9, but it still crashes.
 
Maryam,
I am sorry to hear that these options didn't work for you ....
real-data LES simulation is way complicated and we don't have many experiences on this issue.
Please keep me updated if you eventually figure out how to make it work. Thanks in advance.
 
You must log in or register to reply here.
Top