Version 11 (modified by cetlod, 5 years ago) (diff)

Note on system simplification

Motivation:

We need to simplify the code for maintenance, robustness and future evolution of the dynamical core. This means fewer options, and a consensus on the default (and thus well validated) set of numerics. The idea is to change the current most used choices from :

linear filtered free surface zps ocean with TVD, and standard iso-neutral diffusion ; levitating sea-ice with virtual salt fluxes

possibly to :

Non linear (vvl), split-explicit (ts) free surface, mixte zps/s-coordinate ocean with high order advection schemes, triad iso-neutral diffusion (Griffies operator) ; embedded multi-category sea-ice (exact fluxes).

And this with a removing of obsolescent features and a cleaning of the code.

Timing: achievement within 2 years from now for NEMO v4.0 (to be ready for CMIP6, and on time for the foreseen evolution of the dynamical core)

Notations:

* core action (supported by all system team)

- subject to discussion

? feasible but all conditions required are not yet met



I. CHANGE in options used by default

* non-linear free surface: by default (WP 2014)

* filtered free surface: remove it as well as all its associated solvers (2014?)

required AGRIF with time-splitting (done in 2013) and deeply tested (WP 2014)

* vertical coordinate : default option mixte zps/s-tilde coordinate (thus with variable volume level), with mixte zps*/s*, mixte z/s, zps, z and s coordinate system as sub-option (same code used) (2014-2015 ?)

!- tracer solved as thickness weighted tracer: D t (e 3t T) (WP 2014)

!- thickness weighted quantities in the outputs (WP2014)

!- no more substitute (domzgr_substitute.h90): all e3 always defined at before, now and after time steps (partly done in 2013, to be achieved in 2014).

!- adapt OFFLINE for tracer to variable volume (2015?)

!- For OFFLINE, compute the grid in the same way than in ONLINE.

* linear ssh option set without optimization (i.e. starting from the non-linear case with minimal changes, not optimal changes), or linear case suppressed ?? We should probably keep a linear case for test cases. (2014-2015 ?)

* sea-ice : embedded sea-ice as default

!- suppress levitating sea ice with virtual salt fluxes. (2014 ?)

!- Suppress levitating sea ice with mass exchange (?) CICE issue…

!- need the management of grounded-ice & land-fast ice (WP 2014)

!- need BDY compatibility with embedded sea-ice …. (who when?)

- tracer advection scheme: diminish their number (?).

!- Typically one module for 2 nd /4 th order and one for a 3 rd order scheme, both associated with an optional flux limiter to ensure positivity. To be discussed

!- Addition of the rotation of the diffusive part of 3rd order scheme (2014-2015)

- vertical physics: simplification versus diversity

!- keep only one (GLS) ? or maintain some diversity?

!- KPP remove unless if-less KPP is implemented (Lemarié et al 2012). Keeping a validated KPP scheme in the system make sense for comparison purposes.

!- suppress or keep the explicit vertical diffusion?



II.OBSOLESCENT features

mains:

* remove LIM2 : only LIM3 remains in the system (WP 2014)

* remove OBC : only BDY remains in the system (=⇒> done in v3.6)

* remove OASIS 3 and 4 : OASIS-MCT instead (WP 2014)

- duplicate isoneutral operator : keep only the triad operator (Griffies' one) ? Tests needed before decision. Those tests are planed in 2014 (NOCS+LOCEAN)

smaller ones

* remove acceleration of convergence (notably rdttra=F(k) =⇒ unique rdt)

* remove Neptune effect : ORCA1 runs shows its inefficiency ( WP 2014 ).

* remove no-slip accurate option (key_noslip_accurate). Module divcur.F90 (=⇒> done in branch 2014/dev_CNRS0_NOC1_LDF​ )

* suppress cross-land advection : this option is hard coded for ORCA_R2_L31 and is nomore used.

* suppress key_traldf_ano and the associated possibility of computing lateral diffusion on tracer using only the departure from a given profile. (=⇒> done in branch 2014/dev_CNRS0_NOC1_LDF​

- Smagorinsky (?) : use 3rd order operator (UBS) instead and remove it ?

- Non-penetrative convection : with its correction and improvement using alpha & beta (work done by L. Brodeau, Stockholm), it seems now better to keep it in the system.

- runoffs : only introduced as lateral flow: remove the option of runoff as surface boundary condition as well as specific treatment in vicinity of river mouths.

- passive tracers : use trasbc routines for passive tracers to add the river inputs of nutrients

I/O

* remove Dimg format (agreement obtained from the DRAKKAR group) (2014)

? remove IOIPSL : 2 issues, still use in some places, and XIOs portability… ( probably much too early to decide that )

* use of XIOS logical to control the computation of diagnostics (2014-2015, already done LIM3, see dev_r4028_CNRS_LIM3_MV2014)

CODE cleaning

* LDF: remove 1D and 2D cases, keep only 3D ; laplacian and bilaplacian coefficients defined at the same points (T and F) ; time varying coefficients (function of growth rate of baroclinic instabilities, of the local velocity…): better integration rather than configuration dependent cases. This is also associated with the re-writing of bilaplacian operator on tracer and on dynamics as re-entrant laplacian operator. As a results, curl and div are no more required in the restart file. On going work, see #1260 and the associated wiki page 1260_CNRS0_NOC1_LDF

* OBS operator : clean the stand-alone implementation (WP 2014)

* avm defined at t-point =⇒ suppress avmu, avmv and ~10 lbc_lnk calls (2015 ?)

* suppress most of the CPP keys, if not all of them (2014-2015 ?)

* dynamical allocation: return back to standard allocation (?) (no more CALL wrk_alloc and CALL wrk_dealloc) Evaluation to decided what to do (WP 2014)

* no more potential density (zdfnpc and MLD computed from alpha & beta) ( WP 2014)

* remove all hard coded specific alterations associated with configurations (ORCA,…) while introducing associated input files (2014-2015 ?)

* same for the definition of coordinate system (s, mixed z-s): only a simple case hard coded. For more sophisticated cases, the required information should be read in an input file. (2015 ?)

- remove the key_esopa option that allow to compile and test all code option that are controlled by a cpp key. With the removal of nearly all cpp keys, it should become useless…