Weird warning in tpindex with 64 levels

[ehouarn]

Running Saturn case with 64 levels, I get this weird warning from tpindex: 000: tpindex: Caution! Pressure of upper levels lower than ref pressure for k-coef: 000: k-coeff fixed for upper levels 000: PWL= -3.11805930589871
000: PREF(1)= -3.00000000000000
But codes seems to run properly

[aymeric-spiga]

Does not seem to occur with 32-level simulations (tried with debug mode)

[aymeric-spiga]

So @ehouarn you have it in 1D or just in 3D?

[ehouarn]

3D. no trace of that message in 1D.

[ehouarn]

After some investigations: The values mentioned in the message are powers of pressure, not pressure itself. So this just means that some information is sought at 10(-3.1)=7.9E-4, outside the predefined pressure grid (which stops at 10-3).

And there is a mismatch between the top layer pressure in 1D or 3D. In 1D we have: PRESNIVS 294147.240516825 278690.364708110 254015.274914879
... 0.279683499458116 0.220006832954346 0.173063504426904
0.136136574316017

Whereas in 3D: 000: PRESNIVS( 1 )= 294147.240000000 Z ~ 0.985100817881494
... 000: PRESNIVS( 62 )= 0.220006830000000 Z ~ 706.281722064773
000: PRESNIVS( 63 )= 0.173063505000000 Z ~ 718.281721227778
000: PRESNIVS( 64 )= 7.619749500000000E-002 Z ~ 759.298222224199

[aymeric-spiga]

So this is only the latest level that is different. I think in 3D fully coupled with the dynamical core the latest vertical level should be closest as possible to zero

[aymeric-spiga]

For what it's worth, the problem disappears (no warning) when running the model in makestart with 61 levels instead of 64 levels (removing the 3 uppermost levels). Also the model is running MUCH faster. Looks like the warning does not make the model crash, but slows it down.

[aymeric-spiga]

I added a new branch https://github.com/aymeric-spiga/dynamico-giant/tree/work_61levels to share my configuration.

[aymeric-spiga]

The problem was that at 64 levels, a level in DYNAMICO is higher than the uppermost limit of the correlated-k database (here 0.001 Pa), hence the warning. It is not seen in 1D calculations because the way the last level is treated differs between DYNAMICO and LMDZ: DYNAMICO has a higher boundary that is always at p=0 to ensure conservation. Using 61 levels solves the problem, but resetting vertical levels in z2sig.def could have done the job as well. Or recomputing correlated-k grid, but here we do not necessarily need to go higher than 0.001 Pa