WRF initializing WRF vs nesting

Questions and other topics related to UEMS 15.
Post Reply
norulz
Posts: 92
Joined: Thu Mar 12, 2015 1:43 pm

WRF initializing WRF vs nesting

Post by norulz » Thu Dec 15, 2016 6:01 pm

Hi all,

I am interested to know if there should be an inherent difference in the simulation output between the two following cases:

Case 1. A single 1 way nested run with gfsp25 initializing a WRF run with outer domain having 12km grid and inner nested domain with 4km grid.

Case 2. First run gfsp25 initializing a WRF 12km run with no nesting and then using output from this run to initialize a second WRF run with 4km grid (no nesting again).

Can I expect the same simulation results from both cases?

David

meteoadriatic
Posts: 1574
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF initializing WRF vs nesting

Post by meteoadriatic » Thu Dec 15, 2016 6:31 pm

Yes,

Because in first case you feed nest domain with boundary data on every time step, and on second case you feed it only in output time step of first domain. First case will always be better and probably faster choice. Second case is useful, in my opinion, only if you don't have enough RAM to fit both domains in, or something similar wierd situation.

norulz
Posts: 92
Joined: Thu Mar 12, 2015 1:43 pm

Re: WRF initializing WRF vs nesting

Post by norulz » Thu Dec 15, 2016 8:29 pm

Hi,

Thanks for the observation. Adding boundary conditions at every time step should definitely provide better simulation. It may, however actually be slower.

My motivation for exploring the option of WRF separate runs stems from looking for better economy in covering much of the Mediterranean with 4km grid. I do this with multiple productions of 12 & 4km nested runs. This results in a lot of overlap of the 12km grids.

I have tried a larger 12km grid with two nested domains of 4km but found this to be too slow for operational production so I split the work over two units.

meteoadriatic
Posts: 1574
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF initializing WRF vs nesting

Post by meteoadriatic » Thu Dec 15, 2016 10:04 pm

Why do you think it is slower? You have same amount of total grid points to calculate in both cases, and in second case you need to postprocess coarse domain (wrfout -> grib) and then ungrib those gribs (those are two additional tasks that does not take place in first example). So theoretically it should be slower not faster than doing simultaneous nesting.

However, you can try and benchmark yourself to ensure which way is faster.

Post Reply