4.8.0 Grell-Freitas-Li scheme kills convection

M

meteoadriatic

Member
Hello,
The version 4.8.0 replaces pretty good scale-aware convection scheme, Grell-Freitas, that was battle-tested, extremely good behaved, with "improved" one, namely Grell_Freitas-Li, but this one shows pretty dramatic regression in quality.

I'm testing one challenging convective case using 3 km grid and just wondered if GFL will do better than GF. So I compiled 4.8.0 and used bit-identical namelist, with cu_physics = 3, which means that active cumulus scheme is now GFL instead of GF.

These are results, which speak for themselves:

1781214393678.png

All coastal convection that scale-aware GF correctly allowed to be explicitelly resolved by 3 km grid, is supressed and turned into drizzle in GFL.

I don't understand why this has been done (replacing good, operational scheme by many users, with new, untested one, instead of adding new option along existing one)?

Best wishes,
Ivan
 
Hi,

That should not result in what we see here. The fact that there is no coupling between convective clouds and radiation, means that radiation is not modified by the convective scheme, which makes small difference yes, but if anything, in oposite direction I think (more convective energy present). What I think the problem is, is that the new scheme is tuned for global simulations and coarse grids and it now behaves almost as classic, non-scale-aware schemes. Which I'm just fine with, but the main question here is - why the old GF is removed. It was perfect cu scheme.
 
Could be that nobody checked how its scale-awareness behaves at 3 km? At that resolution you're already below where a CU scheme should do much anyway. I guess the target range is more like 7–12 km, where the scheme has to decide between parameterized and explicit convection.


Did you try it at something coarser (9 km for instance) and check how it compares to the old GF scheme there?


Either way your main point holds: GF did the right thing at 3 km by handing convection to the grid, and if GFL won't, that's a regression regardless of grid spacing. Perhaps you should consider filing a GitHub issue.
 
@ meteoadriatic

Thank you for reporting this issue. I also have troubles to implement the GFL scheme for 15km resolution runs at the right beginning, but haven't found time to further investigate what is wrong.

Would you please upload your namelist.input and namelist.wps for me to take a look? It is also helpful to let me know what data you used to drive this case. I would like to repeat your case, and explore what is going on.

Thanks.
 
Hello Ming

To be honest that was a just quick side-test, I don't have particular namelists anymore but I guess any will produce the similar result. Input data was ERA5. What I remember is that PBL was YSU, surface layer Revised MM5, Radiation Goddard, LSM is NOAH-LSM, microphysics WSM7. Other small details I can't remember but probably do not matter.
 
You must log in or register to reply here.
Back
Top