[MITgcm-devel] verification experiment for pkg/bbl

[Menemenlis, Dimitris (3248)]

Oops.  Sorry about the 9pm (PDT) check-in.

I found out cause of high sensitivity of bbl verification experiment.
There is a relaxation time scale in pkg/bbl, which defaults to 24 hours,
and which is too close to the 12-hour tracer time step of  global_with_exf/input.yearly

I have replaced this time scale with 10 days, and now the experiment is much less sensitive:

Output file created on baudelaire:
Y Y Y Y>16<16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 pass  global_with_exf.yearly

On my mac laptop I now get:
Y Y Y Y>14<13 16 16 14 16 16 14 13 16 13 14 13 13 14 13 16 pass  global_with_exf.yearly

which is similar level of accuracy as for the default global_with_exf experiment:
Y Y Y Y>14<13 16 16 16 16 16 14 13 16 13 13 14 12 13 16 16 pass  global_with_exf

Dimitris Menemenlis

On Apr 2, 2012, at 8:27 AM, Jean-Michel Campin wrote:

> Hi Dimitris,
> 
> We will see tomorrow how it goes through the different 
> platforms/compilers.
> I take this opportunity to remember everyone (but specially the 
> West coast developers) that the daily tar file + most of the 
> automatic testreport that are run over night download the code
> between midnight and 1.am, so that, in order to avoid confusion,
> it's better to avoid checking-in changes to MITgcm during that
> time.
> 
> Cheers,
> Jean-Michel