[cam-users] CAM blow-up

[David Noone]

Hi Ragoth,

Smaller time step is probably the best option - the topology of the
numerics requires it. But there are always another ways. You could try
ramping up the diffusion coefficients (horizontal, but esp. vertical).
This will artificially (and probably non-physically) smooth the wind
(etc.) fields and reduce the Courant number. This smoothing is
tantamount to reducing the resolution, and you're essentially back where
you started (i.e., don't be tempted by this). However, if you feel that
the blow-up is related to inaccurate estimation of the vertical velocity
field (from integration of the continuity equation), this might be a
useful tool.

Another option is to use a semi-Lagrangian algorithm for stepping the
dynamics - at least one is provided with CAM. Semi-Lagrangian schemes
are not limited by Courant condition for stability. Not strictly a
unique feature of semi-Lagrangian scheme is the ability to decompose the
transport into a large part which is known analytically, and a smaller
residual which is found numerically. In the case of CAM's interpolating
"SLD" core, this is all done by interpolation. I understand the finite
volume core also make use of this Lagrangian property, but haven't used
it myself. Might be worth a shot, although you should be away that the
results will differ from that of the spectral Eulerian core and you
should take care in verifying the science. Once again, your problem
requires increased accuracy in time, and a smaller time step would be my
first choice (reduction to 1200 seconds isn't too much of a hit!).

Good luck,

David


On Tue, 15 Jul 2003, Ragoth Sundararajan wrote:

> hi,
>
> I am running CAM at a higher vertical resolution. This worked for several
> runs so far, but the present run seems to blow up in the following way:
>
> NSTEP =   32121   8.935174808223684E-05  7.179735399927720E-06
> 250.900  9.84382E+04  2.204584220356811E+01 0.72  1.13
>   NSTEP =   32122   8.937561785090568E-05  7.387377207262184E-06
> 250.899  9.84382E+04  2.204539090829577E+01 0.75  1.11
>   NSTEP =   32123   8.942472814784718E-05  7.950277355457169E-06
> 250.899  9.84382E+04  2.204453809698432E+01 0.90  1.26
>   NSTEP =   32124   8.953576120359026E-05  9.358443771738252E-06
> 250.899  9.84382E+04  2.204432375119116E+01 1.16  1.48
>   SPHDEP: ****** MODEL IS BLOWING UP *********
>
>
> Parcel associated with longitude    60, level    37 and latitude    49 is
> outside the model domain.
>   ENDRUN IS BEING CALLED
>   CUBYDR: Departure point out of bounds: jcen,icount,nlon*plev= 53 6015 6016
>   ENDRUN IS BEING CALLED
>
> I see that the courant number (vertical) is consistently above unity in the
> last several time steps, and when the horizontal courant number also
> catches up to unity, the model blows up (It's strange the way this number
> jumps up to unity in a couple of time steps). Does this mean, reducing the
> time step is the only solution ? I have run a few other cases with higher
> vertical resolution with 1200 sec time step, and they worked (although
> courant number violation occurred there also, but no blow up).
>
> Any ideas will be appreciated. Thanks in advance.
>
> Ragoth Sundararajan.
> MIT.
>
> _______________________________________________
> cam-users mailing list
> cam-users@ucar.edu
> http://mailman.ucar.edu/mailman/listinfo/cam-users
>
>

-- 

--------------------------------------------------------------------
California Institute of Technology, Pasadena, CA, 91125, USA
Division of Geological and Planetary Sciences
Mail Stop 100-23, Phone (626) 395-6982
http://www.gps.caltech.edu/~dcn