Transition to Thompson Microphysics Scheme for Microwave All-sky Assimilation #719
Assignees
Milestone
Comments
6 tasks
|
@azadeh-gh and @emilyhcliu : what's the status of this issue? |
|
@RussTreadon-NOAA The 4 weeks cycled experiment run for v17 and v17+ThompsonMP modification in GSI are completed. I'm going to add the PyGSI evaluation plots and link to MetPlus plots here. |
|
Thank you @azadeh-gh for the update. To which branch have the GSI changes been committed? This branch should be linked to this issue. Since we are running parallels, we should open a draft GSI PR. Can this be done? |
6 tasks
|
@RussTreadon-NOAA Yes, draft PR#743 |
|
Perfect! Thank you @azadeh-gh for opening draft PR #743. Who would you like me to assign as reviewers? |
|
I can be the reviewer. |
|
@RussTreadon-NOAA I think @emilyhcliu @ADCollard @xincjin-NOAA ? |
|
@emilyhcliu and @xincjin-NOAA have been added as reviewers to PR #743 |
|
WCOSS2 ctests The The initial radiance penalties differ between contrl ( The |
Looking at the code changes. I am looking into crtm_interface.f90 .... |
|
Ran mp_physics = 11 (GFDL) using GSI develop (ctl) and GSI develop+Thompson (exp) with the same initial conditions and namelist options. Checking 1st out loop results ctl Looking into details in J table for exp and ctl ... exp ctl Will find out why there are differences in atms n20 tomorrow. |
|
Thank you @emilyhcliu for digging into the differences. The behavior in your test is consistent the The contrl and updat initial radiance penalties are identical for the first 8 printed digits. |
|
@emilyhcliu and @azadeh-gh : please do not misread my comments about analysis differences. Differences are OK as long as they are expected and/or explainable (documented). If differences are not expected or explainable, we need to figure out what's going on and document it. Do ctests on other machines yield results consistent with WCOSS2? |
|
@azadeh-gh I suggest using the single-cycle run to do sanity checks:
The expected results are:
|
|
@emilyhcliu Thank you Emily. I will run these experiments and update you. |
|
@azadeh-gh I created a branch feature/thompson-emily from GSI develop and manually added changes from your feature/ThompsonMP_Azadeh This branch (feature/thompson-emily) can reproduce GSI develop result. Please take a look and compare with your and run single cycle tests scripts: single-cycle runs: /scratch2/NCEPDEV/stmp1/Emily.Liu/GSI382
The results of the three runs are identical |
|
@azadeh-gh Here is the next step.
And then run single-cycle test with imp_physics == 8 I will have this done this afternoon. |
|
@emilyhcliu Thank you Emily, I added your changes to the code. The results of the single cycle run from GSI develop branch and GSI feature/ThompsonMP_Azadeh branch with GFDL physics (imp_physics==11) are identical. |
|
WCOSS2 test
This test is not a reliable measure memory used by |
|
@azadeh-gh I ran 2023050200 and 2023050206 for all AMTS and AMSU-A without thinning and set use_edge to true. So, we should get good coverage to check the TCC from GFDL and Thompson. The output diagnostic files can be found in the following location: Please let me know if you have any questions. |
|
@azadeh-gh We fixed the bug in the effective radius for Graupel last Thursday by changing 10E4 to 10. We can activate your GSI PR for active review after double-checking the following:
|
|
@emilyhcliu Thanks Emily. The reason I changed the Graupel fix value from 10E4 to 10 was the fix value in fv3-jedi code. The units of Graupel equation in the slides you shared with me before were different from fv3-jedi. |
|
TCC plots for Thompson vs GFDL
|
|
|
|
For debugging: Search the keyword |
|
Based on the cloud fraction formula (Xu ad Randall), there are some sanity checks we can do: The following MPI processor (1013902) contains many cloudy profiles. Here is the first profile from that MPI output. |
|
@emilyhcliu The routine calculating cloud properties (paths and cloud fraction) used with Thompson is named progcld_thompson_wsm6 in radiation_clouds.f. It calls cloud_fraction_mass_flx_2 to calculate the cloud fraction for radiation. The subroutine that I added to the crtm_interface code is similar to cloud_fraction_mass_flx_2. |
|
Hera ctests results
|
I set lmfdee2 = .true. for cloud fraction calculation in the crtm_interface.f90 (They are still running.....) |
Results from adding lmfdee2 = .true. for cloud fraction calculation in the crtm_interface.f90
|
adding lmfdee2 = .true. for cloud fraction calculation in the crtm_interface.f90 increased cloud fraction
|
|
@azadeh-gh Could you also plot ATMS_N20? Thanks. |
|
@emilyhcliu ATMS_N20 plots using lmfdee2 = .true. for cloud fraction calculation |
|
allskyens_ctrl : imp_physics=11 in GSI
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
PR#743
Description:
Transition from using the GFDL microphysics scheme, currently in operation, to the Thompson scheme (GFSv17) is proposed for the microwave all-sky assimilation. This transition aims to enhance the accuracy and reliability of forecasting.
steps:
1. Read in Hydrometer Variables:
Read in two additional hydrometer variables (rain and ice number concentrations) from the 6-hour forecast model.
2. Modify CRTM Interface:
Modify the Community Radiative Transfer Model (CRTM) interface to include the aforementioned additional hydrometer variables before calling CRTM.
3. Examine Simulated Brightness Temperature:
Examine the simulated brightness temperature after implementing the modifications mentioned above, utilizing diagnostic files for analysis.
4. Run Cycled Parallel Experiment:
Conduct a cycled parallel experiment to assess the impact of replacing the GFDL microphysics scheme with the Thompson microphysics scheme on the global forecast skills.
The text was updated successfully, but these errors were encountered: