Version 2 (modified by jchanut, 3 years ago) (diff)

REPORT OF THE WORKING GROUP MEETING (January 30, 2014)

Participants : Rachid Benshila, Daley Calvert, Julien Calandreau, Jérome Chanut, Laurent Debreu, Christian Ethe, Rachel Furner, Claire Levy, Gurvan Madec, Sébastien Masson

1) Discuss the update of the 3.6 version

History :

  • AGRIF version (and corresponding NEMO subroutines) has been updated by Julien Calandreau (INRIA) and Laurent Debreu
  • Jérome Chanut has made a lot of improvements and bug fixes in his Gilbraltar Strait configuration
  • Same for Sébastien Masson in the Pulsation ANR project
  • Other developments in LIM, PISCES

Bugs that have been identified and corrected since version 3.4 :

  • zwork array (J. Chanut): a global array « zwork » was defined in the libmpp module and (due to a bug in the AGRIF converter) this led to several problems and in particular the update step was not performed …
  • Diffusion coefficients in the corners of the sponge layern were wrong (J. Chanut)
  • Scale factors were not correctly computed in the ghost cells (where it is assumed that we are in full steps) (J. Chanut)

→ same problem for OBC

  • TKE (S. Masson): near the boundaries, spatial averaging operations led to wrong Kz in the first cell inside the high resolution domain. Extrapolation is possible but does not fulfill the requirement that with, a mesh refinement factor of 1, the obtained solution of the HR grid is identical to a solution computed without a zoom. S. Masson has implemented exchanges of TKE between coarse and fine grids.

→ Same problem for GLS and for OBC

  • Asselin Filter (J. Chanut) : when updating the « after » (n+1) fields, the corresponding corrections of the « now » fields were not done.
  • arrays out of bound in procnames in MPI (wrong declarations of AGRIF profiles in agrif_declare_variables: jpi, jpj have been replaced by ncli,nlcj)

Bugs / Problems that have been identified

· The bathymetry connection check is not complete (add a check that the coarse grid bathy have been correctly updated)

· Reproducibility is not guaranteed (due to MPI problem) when ln_spc_dyn is set to true in namelist (which means that we take 0 as special value for dynamics in interpolations/updates).

· Recursive updates. When more than 1 zoom level are set, the update should start from the highest resolution grid and go recursively to the root grid. Laurent implemented changes that makes this possible.

  • Use of AGRIF for degradation online (problème de north pole, coarsening)
  • Average operators with update values
  • High order operators with coastlines
  • BDY problem have been solved

Improvements

  • AGRIF and time splitting (OK for version 3.6, end of the year, Jérome)
    • only consider forward integration of the forcing mode permits to ensure volume conservation
    • without update of the parent grid right hand side
  • Performance of parallel runs
    • allows the distribution of the processors among the fine grids

Rachid Benshila is ready be in charge of the merge

2) What are the known open problems ? (not to be solved for 3.6)

  • Number of ghost cells (L. Debreu)
    • Antartic problem (periodicity): require only one ghost cell instead of 2
    • Number of required ghost cells for high order advection schemes may be larger than 2
    • → Remove hardcoding of the number of ghost cells, do the required modifications to the AGRIF “profiles” (Laurent)
  • North fold problem
  • Timings for each grid

3) What are the main next developments?

  • Coupling between grids at the level of the barotropic mode (in time splitting version) (Jérome)
    • requires some modification in the organization of the step routine of NEMO
  • Vertical grid refinement (Laurent)
  • Online coarsening for tracers using AGRIF (L. Debreu, G. Madec)
  • AGRIF with variable volume

4) Nesting tools:

Two stages:

a) for 3.6 version

  • agree on all the points: bathymetry smoothing, update of coarse grids (take the maximum as default)
  • Incorporate corrections done by Jean Marc
  • The nesting tools are only essential for the creation of bathymetry:
    • forcings can be interpolated on the fly
    • → Do a list available tools for 3D interpolation (initialization / restart fields)
    • For 3.6 keep the 3D fields interpolation in nesting tools

b) next version

  • The definition of vertical coordinates should be done outside NEMO
  • Nesting tools for the s coordinates

5) Documentation:

  • Update of the nesting tools 3.6 documentation
  • Update of the users guide: describes what is possible or not with AGRIF.
  • Documentation for nesting in NEMO/AGRIF: starts from the notes of Sebastien, then circulate the powerpoint
  • GMD paper

6) Standard configuration and sanity tests

The standard configuration should:

  • Include more than one level of embedding (test of the recursive update starting from the finer resolution grid(s))
  • Uses time splitting
  • Include a zoom in a sea ice area (Labrador Sea, LIM3)
  • Include a zoom with PISCES
  • Zoom in Perou Chili

Sanitiy tests

  • Zoom with a mesh refinement of 1
  • Test constancy preservation for tracers

Longer visions of having AGRIF zoom in AMM

S coordinates zoom inside a z coordinate parent grid.