[Wrf-users] Is it a must to run wrf with a nest

[mmkamal at uwaterloo.ca]

Hi there,

I would like to offer the following seven points (the first two copied  
from "WRF Advanced Usage and Best Practices" lecture note):

1) Nestng is probably needed if your target resolution is much less  
than your analysis resolution.

2) Outer domain grid size could be 1/3 analysis (or boundary data) resolution.

3) In your case, analysis data set's horizontal resolution is 83 km.  
Therefore, according to suggestion 2, you can set your outermost  
domain's grid spacing (83/3= ~ 27 km). Following a 1:3 parent:daughter  
grid ration, your second domain's resolution would be 9 km; 3rd  
domain's grid spacing would be 3 km and 4th domain's grid spacing  
would be 1 km.

4) WRF run times depends on multiple factors, including grid spacing,  
integration time steps, choice of physical parameterization schemes  
(e.g., Morrison double moment microphysics schemes takes nearly 15%  
more computing time than WRF single moment six class microphysics  
scheme), the number of grid points WRF needs to integrate (i.e., if  
you have 200 grid boxes along east-west, 175 north-south, and 28  
vertical layers then WRF needs to integrate 200*175*28 = XXXXXXX  
points. For second nest, the number of grid points would be:  
3*(east-west grid box number * north-south gridbox number * vertical  
layers number). For 2nd nest number of grid points would be:  
9*(east-west grid box number * north-south gridbox number * vertical  
layers number). Now, you can imagine how many grid points you need to  
integrate for each successive nest. Model integration time will follow  
nearly the same trend. Nest will take approximately four times higher  
CPU time than parent domain.

5) Now, depending on your computing resources (which is not too much)  
you can consider only one nest. In that case your highest resolution  
domain would be 9km (if you set your outer domain grid spacing 27 km).

6) You can set more than one nest if you set a small domain area for  
nest compared to parent domain. You can not set larger domain area for  
outermost domain if you would like to use more than one nest.

7) So, while running WRF model, one should follow the famous proverb  
"cut your coat according to your cloth".


Cheers,
Kamal





Quoting afwande juliet <afwandej965 at gmail.com>:

> Thanks guys for this useful piece of information. I have 32cpus and 2
> nodes. I am testing many convective schemes over East Africa. If I got you
> right, if my data is at 83km, then the outer nest should be between 5-3
> times less, that means the inner nest will be about 9.33km if ratio 1.3 is
> considered. My confusion is 9km is too expensive to run than 28 or 32km.
> For my case do I set the domain such that the inner nest is 28km or what??.
> This is because the if outer nest is 32km then inner must be 10.5km there
> about which is too expensive.
>
>
> On Jul 25, 2016 9:05 PM, "Mike Dvorak" <mike at sailtactics.com> wrote:
>
>> Hi there,
>>
>> You typically don't have to start with your first domain at the same
>> resolution of the input dataset. I would start out with something on the
>> order of 3 to 5 times smaller in delta x and y. So if your input data is 83
>> km, your WRF parent (first) domain could be roughly between 28 to 16 km.
>>
>> You might run some sensitivity experiments to see how well the model
>> matches observations by going directly to 16 km. The results might be good
>> enough for your purposes and would save you having to run that extra
>> domain. Removing parent domains does *not* often make a huge difference
>> in wallclock time because the inner nests run multiple timesteps for each
>> time the parent domains runs.
>>
>> Best of luck with your experiment.
>>
>> Cheers,
>> Mike
>>
>> On 07/24/2016 09:34 AM, afwande juliet wrote:
>>
>> Dear users
>> I am running wrf using era interim which is at 83km resolution. I am
>> downscaling to about 16KM nest
>> It's very slow. I even deactivated the parent domain so that I only get
>> output from nest to reduce runtime but still takes 84hrs to run 3months
>> simulation which is too slow.
>>
>> What would be the danger if I run without nest to reduce runtime
>>
>> Would there be any mismatch between the driving data and the model
>>
>>
>> _______________________________________________
>> Wrf-users mailing  
>> listWrf-users at ucar.eduhttp://mailman.ucar.edu/mailman/listinfo/wrf-users
>>
>>
>> --
>> [image: Sail Tactics logo]
>> Mike Dvorak, PhD
>> Founder
>> Sail Tactics, LLC
>> Corpus Christi, TX
>> +1 650-454-5243
>> http://sailtactics.com
>>
>> _______________________________________________
>> Wrf-users mailing list
>> Wrf-users at ucar.edu
>> http://mailman.ucar.edu/mailman/listinfo/wrf-users
>>
>>
>