Monday, July 7, 2008

back to modeling

The return to modeling. I have worked extensively with cumulus parameterization schemes in mesoscale and now regional climate models. No matter how good they do, there are serious known limitations to the physical schemes. The problem is that most people do not do a very good analysis on the known unknowns (the things we know we don't know) or at least the characteristics that have not been examined yet.

Cumulus schemes transport heat and moisture (vapor, microphysical species, and rain) from the lower levels to the upper levels of the atmosphere. They do not typically involve momentum transfer though there are always exceptions.

Cumulus schemes do NOT have:
1. sufficient variability to handle tropical or intense midlatitude convective heating,
2. sufficient resolution to handle MCS stratiform cloud shields including the associated coupled heating-cooling vertical dipole,
3. universal trigger function that works in the topics and midlatitudes,
4. can not generate significant heavy rainfall unless numerous non-physical scheme activations occur.

Furthermore, my work clearly shows that PBL temperature and moisture biases exist depsite resolution (4 vs 8-50 km). Vertical velocity is DAMPED when a cumulus scheme is used, not because the instability is being removed but because the cumulus scheme barely counters the destabilization, thus the vertical velocity distribution does not expand.

The cumulus scheme's are heavily untested at fine grid spacing. Good simulations with a cumulus scheme do not mean the cumulus scheme was either the cause of success or failure.
I think we need to ask the following:
1. What unique properties of a cumulus scheme contribute to the succesful components of a good forecast?
2. How does the cumulus scheme interact with other physics?
3. How does a change to the cumulus scheme impact the simulation?
4. What is the tendency of a particular series of changes to a cumulus scheme?

I would also urge that anyone using a cumulus scheme that wants to nest to fine resolution: use it to drive a seperate model run. You dont want cumulus tendencies going through multiple domains in a nested two-ay or one-way feedback.

Nesting in my opinion is not a proven concept until some more work on physics is completed. Early nesting used simple CPSs on the outer coarse grid. CPSs have been known to corrupt daughter nests. User beware.