xml_review

Xml review

Use of multiple file mode (allow xios compiled with netcdf_seq, 3 times faster!)
Use of compression level=5 for all variables : check impact on the rebuild time !
sync_freq="1mo" ... may be good to reduce to 5d...
All netcdf variable name as in DRAKKAR :).
In this document, I refer to the variables by their id defined in nemo and field_def.xml

2D fields:
- gridT: sst, sss, ssh
- gridU : ssu, utau
- gridV : ssv, vtau
- flxT : qt_oce, qsr_oce, empmr, saltflx, erp : OK but double check saltflx in case of vvl ? after one output check if the variables are well masked. Correct if necessary.
3D fields
- none in the spinup phase.

3D fields
- gridT : e3t, toce, soce . OK e3t required by vvl. toce, soce are weighted average.
- gridU : uoce. e3u missing. Not sure it is a good idea to skip this metric. I know they are not kept into the restart files (contrary to e3t), but I think that getting e3u is not strait forward from e3t. It is easy in NEMO because many variables are in core-memory. It is not the case off-line. So far, without tide, I agree that diags using e3u_0 should be enough.
- gridV : voce. e3v missing : same comment than for gridU
- gridW : woce. e3w missing : same comment than for gridU. Note that woce is not a weight-average here. (Surface of the model cell is constant in time, therefore the vertical transport can be accuratly computed). Missing: In general we also save vertical diffusion coefficient, it might be of interest also in the spinup phase.

icemod : iceconc, icevolu, snowvol. I think that this is the very minimum. Ice velocities should be important too, and may be some ice fluxes. On the other hand, also check that all these variables are well masked, and correspond to what we want. I notice that no information on ice-categories are saved ...
- I use to call LIM3 output with the suffix icemod3 in order not to clearly separate LIM2 and LIM3 output.