[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: New Abiotic-HOWTO -> Control Runs


> > >4. If you have respected the Equilibrium drift criteria and choose not to
> > >   submit Historical Control and Future Control output, you MUST still
> > >   provide an indication of the drift of your model. In other words, you
> > >   must use the three *ctrlH* output routines (see Abiotic HOWTO) to
> > >   provide your output for another year. For instance, you could provide
> > >   output from the 0-D, 2-D, and 3-D *ctrlH* routines for the year 1775.
> > >   We would then compute your model drift and treat it as constant.
> > >
> > We have found that computing drift rates by grid point based on short
> > runs gives an inadequate indication of drift because there is some
> > interannual variability within the model. It is difficult to separate
> > interannual variability from model drift from two closely spaced
> > snapshots. If the two incidents of output are far apart (say more
> > than a century), then the drift will likely be large relative to the
> > variability, but a decade will not give a good estimate of grid-point
> > drift in many models.
> >
> > I think control runs are essential and will do them. By the way, it
> > took us something over 4000 surface years (with deep ocean
> > acceleration) to get something reasonably converged.

>Just out of curiosity, your 4000 surface years translates to how many
>years in the deepest layer?

We accidently had a lower acceleration during the first part of the 
spin-up, and then increased it when we discovered our error. So, I 
think the total effective time in the deepest layer was about 25,000 
years. However in the first part of the spin-up, acceleration only 
began at around 3 km (going up to 5x in the deepest layer). In the 
second half of the spin-up acceleration started at around 2 km and 
increased to 7.5x in the deepest layer. This is on top of having 
momentum fields evolve more slowly than tracer fields [time-step 
splitting], which allows a longer tracer time step [1 day].

It's all dirty business, but you have to get to a solution within 
some reasonable amount of time.



PS. By the way, we are also trying to do the OCMIP runs using LANL's 
POP model configured to be similar to NCAR's NCOM model. In this case 
we are in the same state as NCAR, where the model is taking forever 
to run because more respect is being paid to model physics.

Ken Caldeira
Climate System Modeling Group
Lawrence Livermore National Laboratory
7000 East Ave., L-103
Livermore CA 94550 USA

tel: (925)  423-4191
fax: (925)  422-6388
e-mail:  kenc@LLNL.gov