Last edited Timestamp?

Author : John Siddorn

ticket : 985

Branch : https://forge.ipsl.jussieu.fr/nemo/browser/branches/2012/dev_r3435_UKMO7_SCOORDS

---

Description

The current stretched sigma coordinate options mean the size of the surface box varies greatly, leading to inconsistent flux inputs. A terrain following coordinate system which still provides a constant surface box height has been developed.

This has been tested in a series of test configurations based upon a shelf model the AMM7 and shown to improve air-sea exchange and reduce pressure gradient errors.

Testing has been done in NEMO3.2 using non-linear free surface and comparing the presently used Song and Haidvogel 1994 stretched with this new formulation. This has been written up in a paper to be submitted shortly (attached).

Coding

The change has meant updates to the following files:

• changed CONFIG/XXXXX/EXP00/namelist -> to include the various options for the coordinate
  - changed NEMO/OPA_SRC/par_AMM_12km.h90 to make it an L50 model required for the stretching
  - changed NEMO/OPA_SRC/DOM/domzgr.F90 -> to include the new stretching function
  - changed NEMO/OPA_SRC/DOM/domwri.F90 -> correct a bug in the _sco write that makes the gdept output 3D

The domzgr has been reordered to make it more readable, with the different stretching functions included as subroutines. The checks on whether the coordinate is sensible has been updated to include looping over mbathy rather than jpk to ensure that when disappearing coordinates are used it doesn't fail when it shouldn't.

Testing

Scientific testing has been done on a combination of the AMM7km configuration and a seamount test case, and is detailed in the attached draft paper. This testing was done in NEMO3.2, so the code has been updated to NEMO3.4 and testing has been done on the AMM12 standard configuration.

Step 1: Test the standard 3.4 code with 50 levels using the Song and Haidvogel stretch. To prevent having to rebuild all boundaries this was done with key_bdy off. Compare with the updated code using the same stretch. This should give near identical results.

Note the test on the coordinates fails with 50 levels in the standard code (although not for 33 levels) - this is a bug, and has been fixed.