Stochastic physics options in SRW-APP

Submitted by junpark-1 on Thu, 08/26/2021 - 21:18
Forums
Short-Range Weather/Convection Allowing Application

Hi, all.

 

I am trying to apply stochastic options (SKEB, SHUM, and SPPT) in SRWAPP v1.0.0 but it didn't work properly.

I employed parameters from 'regional_workflow' and my domain is 300 by 300 with 3km (ESG grid).
My physics suite is 'RRFS_v1_alpha'

Are there additional necessary considerations?

 

Thanks,

Permalink

Submitted by jbeck on Thu, 08/26/2021 - 22:00

Hi Jun,

Can you please provide more information?  How did it not work properly?  Thanks.

Permalink

Submitted by junpark-1 on Thu, 08/26/2021 - 22:18

Hi, Jeff.

 

It did not produce acceptable outputs. I have no issue when I turn off stochastic options.

When I checked 1-minute outputs, it has a very weird signature in the domain.

Here is the 'pressfc' fied in the phyf00:01:00.nc file.

pressfc

I also attached my 'input.nml' file

 

Thanks,

Jun


 

Attach Files
inputnml.pdf
Permalink

Submitted by jbeck on Thu, 08/26/2021 - 22:39

Thanks, Jun.  That is indeed very strange.  We haven't seen that behavior with the CONUS predefined domains.  Can you try your simulation using one of those domains to see if you get the same problem?  Does your 300 x 300 run crash during integration?

Permalink

Submitted by junpark-1 on Fri, 08/27/2021 - 16:20

Hi, Jeff.

 

I have tried 'RRFS_CONUS_13km' pre-defined domain as you suggested with 'FV3_GFS_v15 physics' suite.

However, it stops after fist time step. I do not have any issues in both my old 300x300 domain and this RRFS_CONUS_13km domain if I turn off stochastic options.

 

I attached input.nml and log file.

 

Thanks,

 

Jun

 

Attach Files
inputnml.pdf
nemsexeolog.pdf
Permalink

Submitted by jbeck on Fri, 08/27/2021 - 23:39

Hi Jun,

Are you running with the SRW App workflow?  If so, can you provide your config.sh script?  One other thing to try is setting sppt_sfclimit=true in the input.nml file.  Finally, it would be a good idea to see if you can run with one stochastic physics scheme first, then add the other two to isolate the problem.

Permalink

Submitted by junpark-1 on Sat, 08/28/2021 - 00:26

Here is the config.sh which I have used 'RRFS_CONUS_13km'

 

I am using regional workflow of UFS-SRW APP v1.0.0.

Actually, I already have tried separately turning on and off for three stochastic schemes but none of them work fine for my case.

I also set sppt_sfclimit options on too.

 

Thanks,

 

 

Attach Files
configsh1.pdf
Permalink

Submitted by jbeck on Mon, 08/30/2021 - 22:06

Hi Jun,

I successfully ran the release branch of the SRW App with SPPT and again with all three ad-hoc stochastic physics schemes (SPPT, SHUM, and SKEB) using the RRFS_CONUS_13km domain on Hera for 12 hours (I used the graduate student test cycle of 2019061500).  If you have access to Hera, the path to the run is here: /scratch2/BMC/fv3lam/beck/FV3-LAM/expt_dirs/test_release_stoch/2019061500.  You can see the ufs-srweather-app directory here:  /scratch2/BMC/fv3lam/beck/FV3-LAM/ufs-srweather-app_release.  Unfortunately, I don't have access to Stampede, so I'm not sure why your configuration is not working.  Hopefully you can compare your configuration with the one I ran on Hera.

 

Permalink

Submitted by junpark-1 on Mon, 08/30/2021 - 22:19

Hi, Jeff.

 

Thanks for the update.

Unfortunately, I can't access to HERA. Anyway I will try to use GST data for the application to see if there is difference.

Thank you very much again.

 

Jun

 

Permalink

Submitted by junpark-1 on Wed, 09/01/2021 - 20:28

I updated this post since I have successfully ran with three stochastic physics options when I employed 'RRFS_baseline' branch of UFS SRW APP.

I recompiled NCEPLIBS since this branch requires updated one.

I am going to compare some codes between 'v1.0.0' and 'RRFS_baseline' codes. Come to think of it, the NCEPLIB library also could be a factor for my failures.

Are there anyone who can explain the main differences between 'v1.0.0' and 'RRFS_baseline'?

 

Thanks,

 

Jun

 

Permalink

Submitted by jbeck on Wed, 10/06/2021 - 18:20

Hi Jun,

I'm glad to hear you were able to run with the three stochastic physics options.  The RRFS_baseline branch is specifically for our RRFS_v1alpha physics suite benchmarking runs on a 3-km CONUS domain, and isn't intended to be used by the community.  It contains updates from the develop branch which may be what you needed to run the stochastic physics correctly.  I would recommend updating to the develop branch which contains the most up-to-date code for the SRW App.  If you'd like more information on the changes that have gone into the develop branch since v1.0.0, you can check out our SRW App training presentations and recorded talks here:

https://dtcenter.org/events/2021/unified-forecast-system-ufs-short-range-weather-srw-application-users-training/agenda

This talk will also be of interest:

https://dtcenter.org/sites/default/files/events/2021/19-20-diffs-release-develop-ufs-app-development-dom.pdf

Permalink

Submitted by junpark-1 on Wed, 10/06/2021 - 18:29

Hi, Jeff.

 

Thanks for the kind and detailed replies.

Actually, I have compiled 'develop' branches after I compiled several libraries using 'hpc-stack.
I am testing 'RRFS_v1alpha' suite but I have trouble in running the executable from 'develop' branch.

I think it seems to be related with the 'input.nml'
The executable from 'RRFS_baseline' branch works when I deleted 'do_sfcperts' from 'input.nml which I employed in SRW APP v1.0.0.
However, it didn't work out for 'ufs_model' of 'develop' branch regardless of using stochastic physics.

Are there any necessary changes in input.nml between 'public v1.0.0' and 'develop' branches?

 

Thanks

 

Jun

 

Permalink

Submitted by jbeck on Tue, 10/12/2021 - 15:31

Hi Jun,

 

The develop branch of regional_workflow should work fine for stochastic physics without having to touch the namelist file.  You shouldn't need to modify anything manually.  What problem are you having?  Is it just with the executable?

Permalink

Submitted by junpark-1 on Thu, 10/21/2021 - 14:39

Hi, Jeff.

 

Thank you so much for your suggestions to use regional_workflow.
When I employed 'input.nml' from 'regional_workflow' in 'develop' branch, it worked without any issues.
I found that some namelist variables in 'input.nml' is dissimilar between 'v.1.0.0' and 'develop' branch.
That caused the model stop at the initial time step.

 

Jun

 

 

  • 66 views

Copyright © 2023. All rights reserved.