Very high model top for WRF

[dnolan]

Hello WRF friends,

I'm trying to simulate gravity waves, caused by convection, going up through the stratosphere, with WRF version 3.9.1.1.

My starting point is the standard em_quarter_ss supercell test case, with some changes: bigger domain, more points, more vertical levels.

And, I move the model top up to 50 km.

If I use the standard "stretch_grid=true" with zs = 0.4, ideal.exe fails to put the model top as high as I want it. It also fails to keep equal vertical grid spacing, as DZ between the top few levels increases drastically, so that the vertical grid spacing near the top becomes 1-2 km.

If figured out that decreasing zs (e.g., zs = 0.2) brings the model levels closer and closer to evenly spaced in the upper levels of the domain, and also allows ideal.exe to get closer to the target model top. Near the top, the numerical spacing in the ZNW levels becomes very small.

However, in this case, the model blows up instantly. A downward moving wave of W with values of 50-100 m/s appears in the first few time steps. Decreasing the time step to very small values does not make a difference, nor does different damping options.

So my questions are:

1) Is there something about making the vertical grid spacing in ZNW very very small that causes the model to fail?

2) Or in general is WRF unable to use vertically even grid spacings at very high altitudes?


Thanks for your time.

Dave Nolan

 

[dnolan]

A few updates:

1) Although the input_sounding for the em_quarter_ss case only goes up to 30 km, I extended it up to 70 km using climatological values.

2) I tried the same test in WRF4.1.3 using the same setup. The result was the same, the model blows up in the first few minutes. Attached shows profiles of W at t = 2 min. WRF 4.1.3 appears to set zs automatically based on the model top.


Dave Nolan

 

[kwerner]

Hi Dave,
I ran this by our physics/dynamics specialist. They said using zs =0.2 is reasonable, as it should be a pressure scale height divided by the top height. The problem looks like a hydrostatic imbalance and it is not clear why ideal would not balance that. It could, perhaps, be related to 32-bit accuracy? Perhaps you could try compiling with double precision to see if that makes a difference.

 

[dnolan]

Thanks for your reply. I'm willing to try compiling with 64 bit. I think I know how to do that, but I'm not 100% sure. Can you tell me what to do (make changes in the configuration file, right)?


Dave Nolan

 

[kwerner]

Dave,

When you configure WRF, use the command:

./configure -r8

and then compile. The "-r8" will compile with double-precision. This works with GNU, Intel, and PGI compilers.

 

[dnolan]

Ok, tried it. It doesn't help. The result is exactly the same.

Do you guys want to try it? All you need is the namelist, and the sounding extended up to 70 km, which are both attached.


Dave

 

[dnolan]

Are you guys giving up on this?


Dave Nolan