Standardize the WRF Simulation with ensemble

[satiz]

Hi,
I'm interested in creating an ensemble of 21 members with same IC to study modification of PBL parameterization on cold front. May I know the appropriate way to carry out ensemble run in WRF-4.3.3? I have come across these cases,
1) Altering the wrfinput file: Adding +/- 1 SD of 3-hour mean of a variable (Temperature, U-wind, v-wind, Specific humidity, Pressure etc) as perturbation to the wrfinput file and run WRF
2) Altering the reanalysis dataset initially: Add same perturbation to dataset before starting WPS.
3) Using rand_perturb in namelist file.
(a) In (1) do I have to add the perturbation to all the domains of wrfinput or is it enough to add perturbation only in the innermost domain?
I think the first two cases have to give the same results but (1) most likely takes less time (no need to run WPS!!) for the 21 ensemble members run. Correct me if i'm wrong. How to perform the third case? Specifically how to add perturbation in terms of std deviation of field with rand_perturb?

Thank You in advance

 

[Ming Chen]

Hi,
I'm interested in creating an ensemble of 21 members with same IC to study modification of PBL parameterization on cold front. May I know the appropriate way to carry out ensemble run in WRF-4.3.3? I have come across these cases,
1) Altering the wrfinput file: Adding +/- 1 SD of 3-hour mean of a variable (Temperature, U-wind, v-wind, Specific humidity, Pressure etc) as perturbation to the wrfinput file and run WRF
2) Altering the reanalysis dataset initially: Add same perturbation to dataset before starting WPS.
3) Using rand_perturb in namelist file.
(a) In (1) do I have to add the perturbation to all the domains of wrfinput or is it enough to add perturbation only in the innermost domain?

I would suggest that you add the perturbation to all domains. This is because it is better to keep parent-child domains consistent with each other.

I think the first two cases have to give the same results but (1) most likely takes less time (no need to run WPS!!) for the 21 ensemble members run. Correct me if i'm wrong. How to perform the third case? Specifically how to add perturbation in terms of std deviation of field with rand_perturb?

Option (1) will affect only the initial conditions, and option (2) will affect both initial and lateral boundary conditions.

Thank You in advance

 

[satiz]

For the problem I mentioned earlier I followed adding the perturbation to a single grid point in the outermost domain. The procedure I followed is - Edit the wrfinput_d01 file from real.exe by changing the value of T (original value +/- X; X=1,2) at point (lon,lat,lev)=(5,5,5) of the outermost domain and run WRF.exe. The results have no difference with the CTRL (no perturbation). I was expecting to have differences in the results of perturbation runs wrt CTRL. May i know why there is no difference?

 

[Ming Chen]

I guess you set feedback on in your namelist.input. Please let me know if I am wrong.
In this case, the values from child domain will be used to replace the values at the same grids of the parent domain. This is possibly the reason that you didn't see differences.

 

[satiz]

Yes, I have turned ON the feedback. But the grid cell where I added perturbation is not a part of the child domain. I have 4 nested domain each having elements 100x100 starting with a resolution of 27km, 9 km, 3 km, 1 km. The grid point where I added the perturbation belongs to d01 (outermost domain) and the grid point is (5,5,5) which is on the lower left corner of d01. So, I thought it should produce a small difference in the results right in this case?

 

[Ming Chen]

Are you saying that the pertubation grids are located within the boundary buffer zone?

 

[satiz]

Yes. That perturbation grid doesn't coincides with any child domain.

 

[Ming Chen]

Would you please clarify whether the perturbation grids are located in lateral boundary buffer zone?

 

[satiz]

May I know what is lateral boundary buffer zone? I'm not quite sure what it is. I have attached an image for your understanding of what I did.

 

[Ming Chen]

The lateral buffer zone refers to the outermost 5 grids of your model domain, where the values are updated based on WRF simulation and large-scale forcing.

 

[Ming Chen]

Please see Chaper 6 of WRF Tech Note.

 

[satiz]

Thank You Ming Chen for the clarification. Since the chosen grid point falls within the relaxation zone, any perturbation added will get updated and will not affect the simulation results. Is that correct?

 

[Ming Chen]

That is exactly what happened.

 

[yuankw]

Hi,
I'm interested in creating an ensemble of 21 members with same IC to study modification of PBL parameterization on cold front. May I know the appropriate way to carry out ensemble run in WRF-4.3.3? I have come across these cases,
1) Altering the wrfinput file: Adding +/- 1 SD of 3-hour mean of a variable (Temperature, U-wind, v-wind, Specific humidity, Pressure etc) as perturbation to the wrfinput file and run WRF
2) Altering the reanalysis dataset initially: Add same perturbation to dataset before starting WPS.
3) Using rand_perturb in namelist file.
(a) In (1) do I have to add the perturbation to all the domains of wrfinput or is it enough to add perturbation only in the innermost domain?
I think the first two cases have to give the same results but (1) most likely takes less time (no need to run WPS!!) for the 21 ensemble members run. Correct me if i'm wrong. How to perform the third case? Specifically how to add perturbation in terms of std deviation of field with rand_perturb?

Thank You in advance

Hi, Satiz,
I am also interested in the perturb technology. May I ask you some details?