interruption of high resolution running

[sd16121]

Dear office,

I ran a winter case with only one domain (250m), but it was always interrupted. Could you please provide some suggestions on running this completely?

Please note that another summer case with same configurations can be run completely.

I attached the namelist files and error file for you.

Best wishes,
Stella

 

[kwerner]

Stella,
What type of input data are you using? Your domain resolution should not be more than about a 7:1 ratio to the input data resolution. I suspect that 250m is much more than that, unless you have access to some very high resolution input. This could certainly cause the running issue. You also have time_step = 3, and for a domain using 250m grid spacing, time_step should be no larger than 1.5 (you can use namelist parameters "time_step_fract_num" and "time_step_fract_den" if you want to use a fractional time_step.

I see that you're running with 16 processors, meaning you should have rsl.error.* files for all 16. You should check all the files to see if there are specific error messages listed in each one, and you should start by checking for cfl errors.

grep cfl rsl.*

If so, this means the model has become unstable, most likely due to the time step that's too large.

 

[sd16121]

Dear Office,

Please confirm that if the ratio(7:1) you mentioned equals ERA5 resolution(31 km)/domain resolution?

If so, when I have nested domains, the domain resolution indicates which domain?

Best wishes,
Stella

 

[kwerner]

Stella,
Yes, that is what I mean, and the domain resolution I refer to is the parent domain.

 

[sd16121]

Dear Colleague,

You mentioned domain resolution should not be more than about a 7:1 ratio to the input data resolution, but I found the information at: https://www2.mmm.ucar.edu/wrf/users/namelist_best_prac_wps.html#parent_grid_ratio
It's best to keep the ratio of input data resolution to the coarsest grid resolution (on your domain) to no more than 1:10
Which version is more suitable for WRF 3.8.1?

By the way, I would like to confirm the ratio issue:
only one domain
Do you mean 0.25km(domain resolution)/31km(input data resolution) should be less equal than 7?

Best wishes,
Stella

 

[kwerner]

Hi Stella,
You technically could probably get away with allowing the ratio to be as large as 10:1, but it's better practice to stay lower than that, which is why I gave a safe ratio of 7:1.

If we are being safe and trying to stay within a 7:1 ratio, I mean that the input data (e.g., GFS data) should not have a resolution larger than 7x the resolution of the first domain. If you need to simulate a domain that uses 0.25 km resolution, you will need to add additional domains around that 0.25 km nested domain so that the ratio is not too large. So if you have input data that is 31 km, and let's say you're using a 3:1 parent_grid_ratio, you could have a domain set-up like:

dx = 6750, 2250, 750, 250

Or you could use a 5:1 parent_grid_ratio:

dx = 6250, 1250, 250

But you should probably only use parent_grid_ratios of 5:1 or 3:1. If you're unfamiliar with the concept of nesting, take a look at the section in chapter 5 of our WRF User's Guide.

 

[sd16121]

Dear Colleague,

Thanks for your kind reply. Are there any differences in the simulated results between the parent_grid_ratios of 5:1 and 3:1?

Looking forward to hearing from you.

Best wishes,
Stella

 

[kwerner]

Stella,
Yes, there will be differences in the simulations with a 3:1 and a 5:1 grid ratio. If you'd like to see the differences, you can run a simple quick test with both and compare the output.