I want to setup a mesoscale weather prediction model (WRF or something else) to work with NCEP/GFS meteorological input data with resolution of 0.25deg. My question is how to choose a proper size for the parent domain, considering the resolution of the input data. In my case GFS resolution is around 20km in mid latitudes, is it necessary for my parent domain to be around this value - for example 18km (or 21km) and then use nested domains to go down to 2km.


Or i can start directly with parent domain size 2km? What will be the effect of this big gap 20km to 2km? Do i need a longer spinup so i can resolve clouds for example, if i choose to start straight with 2km resolution? Am i going to miss some kinetic energy from the coarse meteorological input because of dissipation not represented properly? I suppose computation time is not much more if i use 18>6>2km compared to starting directly with 2km. From my experience with WRF, finer domains take most of the time to be computed -ar. 85% in my specific setup.

  • $\begingroup$ there is no hard and fast rule AFAIK. You need to do a lot of trials for your domains. How large is your parent domain ? What is the purpose of your study ? Are you publishing a paper or you working for a meteorological agency ? $\endgroup$
    – user1066
    Commented Jul 3, 2017 at 9:56
  • $\begingroup$ you can keep the size of the finer domain to be very small if you are running high resolution simulations(less than 2 km). But your problem is not likely to be the resolution of the parent domain. It is going to be identifying the lateral boundaries and how that will affect the quality of the forecast. $\endgroup$
    – user1066
    Commented Jul 3, 2017 at 12:25
  • $\begingroup$ The size of my finer domain is 300x200 grid points with resolution of 2km, but my question is do i need to start from domain with18>6km resolution, covering fully the finner domain, or i can start directly with the 2km finer domain.If my meteorological input was with resolution of 4-8km i would go directly to finer domain, but as i am planing to use GFS 0.25deg input, i think its better to use nested domains starting with 18km resolution. $\endgroup$
    – geo
    Commented Jul 4, 2017 at 11:26
  • $\begingroup$ As I said the resolution is a secondary issue. Think of the lateral boundaries and where they will be located. How will that affect your forecast skill ? $\endgroup$
    – user1066
    Commented Jul 4, 2017 at 12:31
  • $\begingroup$ There was a similar question on year ago. Your question is no real duplicate. But the other question's answer might help you: earthscience.stackexchange.com/questions/8067/… $\endgroup$ Commented Jul 5, 2017 at 9:42

1 Answer 1


A good rule of thumb for down-scaling with WRF is not to interpolate down more than 3 or 4 times; with 3 being the standard recommendation. So while 18 km is probably overkill for an outer domain (this is a personal opinion, there are some operational models like HWRF that start at 18km), 2km would definitely be too fine for a parent domain. The main reason for this is the boundary conditions: your wrfbdy_d01 file is interpolated from the original data down to your model's resolution, so for a 2km domain you will be constantly having very low resolution features imposed upon your boundaries, which will cause fairly unphysical conditions at the boundaries throughout your forecast period. By having a wider, lower-resolution "buffer" domain between you and the outer boundary conditions you avoid unphysical artifacts entering your domain.

There are 5 overlap gridpoints at the edge of each domain, so your inner domain should be at least 5 (outer) gridpoints away from the outer domain.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.