real.exe for long-term simulations

[chenghao]

I am running a continental-scale long-term WRF simulation on Derecho, and I have encountered a potential bottleneck with real.exe. Running real.exe with my settings will likely exceed the 12-hour wall time limit.

I've already reduced the update frequency of BCs from hourly to every 3 hours and adjusted the integration time step. I have also tried to increase the number of cores, but to my knowledge real.exe does not benefit much from parallel computing. And I don't think there is a restart option for real.exe.

Does anyone have suggestions on how to address this issue? Thanks!

 

[kwerner]

Hi,
Yes, I believe there is an option for this issue. You can split the real.exe run. For e.g., let's say you are running for the full year of 2023. Then you would do something like:

1) Run real.exe from 1/1/2023 - 6/30/2023
You will get wrfbdy_d01 and wrfinput_d0* (depending on how many domains you are using) files. Save those in a different directory, or as a different name, to keep them from being overwritten.
2) Run real.exe from 7/1/2023 - 12/31/2023
You will get output files with the same names as above, but these will be for the second half of the year. Again, save those elsewhere (or as a different name).
3) Run wrf.exe for however long you wish to run at one time (up to the end of June), but make sure to output restart files at an interval that would ensure that you will have one for 7/1/2023.
4) Run wrf.exe as a restart (restart = .true.) starting at 7/1/2023, using your wrfrst* file from that time, along with the wrfbdy_d01 file for that time.

You can cater the example to the specific dates you're interested in.

 

[chenghao]

Got it. Thanks a lot for the detailed instructions!