Nudging coefficients from NetCDF file

[rsignell]

To create a better initialization field from our Adriatic CTD data, I was thinking about taking the OA'd field, and running the model forward for a couple of days with the observed (weak) wind forcing but strongly nudging back to the initial field along the sections of initial CTD data.

The reason is that while we had dense CTD coverage in the Northern Adriatic, we only cut through the mid and south adriatic pits with one detailed section each (sort of a "feature oriented") approach to sampling. So I though if we assimilate hard to these sections, the density structure from these sections will be advected by the geostrophic flow around the basins and come up with something more reasonable than that obtained by simply using OA.

But to assimilate strongest at the data sections seems to require setting the nudging coefs with short time scales at the CTD locations, and large time scales elsewhere. I thought I could set this in a NetCDF file, but it looks like right now it's only possible to do analytically in set_nudgcof.F.

Should I hack reading some file into set_nudgcof.F, or is there a better way?

Thanks, Rich

Richard P. Signell
signell@saclantc.nato.int

[arango]

Actually, you can assimilate almost anything. It just depends on what assimilation scheme you use. By the way, we have been working for the last 5 weeks on the 4DVAR scheme for ROMS. Everything is code and seems to be working. We still need to make sure that the descent algorithm is correct. This was a very tricky thing to code. We are very happy with our progress. Now, we are working on the twin experiments.

You don't need to do nothing in "set_nudgcof.F". I'll use the OA error which is normalized by the field variance. This error then is nondimensional and a function of your data coverage and choice of spatial and temporal decorrelation scales. This error field is read in ROMS (see get_data.F) and converted to nudging time scales for assimilation via nudging. For instance, if you select a nudging scale of half a day, this scale will be associated with (1-error). That is, very strong nudging where there is data and no nudging where there is no data (when there is a large OA error).

Do this makes sense?

Good luck, H

Hernan G. Arango
arango@imcs.rutgers.edu