• 2.1 Merging  LPC_TRAJ  and  LFULLIMP  code
• 2.2 Retained functionalities for harmonized schemes
• 2.3 Restricting functionalities
• 2.4 Link with physics

2. Harmonization of iterative schemes

2.1 Merging  LPC_TRAJ  and  LFULLIMP  code

The LPC_TRAJ  and   LFULLIMP  schemes are based on exactly the same scientific idea, but implemented in the code through a different algorithmic way (i.e. incremental form for LFULLIMP  and non-incremental form for   LPC_TRAJ ). Rewriting   LFULLIMP  in the normal non-incremental form should normally not change the response of the scheme, and should require very few changes in the code. This statement has been carefully examined between all persons who have been involved in the development of these schemes, and has been agreed as true. This transformation will thus be tested first, since it constitutes the first necessary condition for a deep harmonization of the  LPC_TRAJ  and  LFULLIMP  schemes. If, for a reason not detected yet, a serious difficulty was to appear while verifying the validity of this statement, then the harmonization between  LPC_TRAJ  and  LFULLIMP  schemes could be only much more superficial. However, this situation has rather few risks to really occur.

If this first test is successful, showing that this transformation from incremental to non-incremental form has effectively no impact on the response, then these two schemes will be deeply merged into a single one, written of course in the non-incremental formalism, since it is the form which has been chosen for the general design of IFS/ARPEGE/Aladin since its origins.

2.2 Retained functionalities for harmonized schemes

It has been agreed that:

• LPC_OLD  should stay in the cycles for the time being.
• LPC_FULL  should also go to the cycles.
• LFULLIMP  and  LPC_TRAJ  should also be merged as stated above.
• the pseudo-second-order decentering (XIDT) should be retained in cycles as well as the first-order decentering for all schemes.
• the possibility to use either high-order (HO) and low-order (LO) interpolations or only HO interpolations should be retained for  LFULLIMP .
• the possibility to have different values of the first-order decentering parameter for some variables should also be kept for  LFULLIMP .

2.3 Restricting functionalities

Inclusion into official cycles also implies that the less important research switches allowing for slight changes in the scheme should be removed prior to inclusion, in order not to increase too abruptly the size of the code for small details.

After many negociations, it has been agreed to retain almost all functionalities currently implemented in iterative schemes. In particular it has been agreed not to eliminate the possibility of using classical time-extrapolation or LSETTLS time-extrapolation for the predictor-step, even the value of this is not obvious a priori (a non-extrapolating scheme for the predictor step is claimed to be sufficient in the literature).

At the beginning, it had been thought that this kind of "almost useless" functionalities would make a large increase in code volume too important, and also that this would handicap the readability of the code. However, after a closer examination it has been decided that thanks to the rationalization of the code (mainly of LATTEX), these functionalities would enter quite naturally in the code, and would not generate any useless complication in the data flow. On the other hand the code for iterative scheme will reflects more its SI counterpart. A consensus was finally obtained on the fact that even with these "not very useful" options, the final code will be much more readable and well written than it is now.

It also has been agreed to eliminate the possibility of having several different values of the pseudo-second-order decentering for the differents iterations, hence to retain only one single value XIDT for the pseudo-second-order decentering, valid for all iterations, and for all variables.

2.4 Link with physics

It has been agreed that the convergence of scientific choices between IFS, ARPEGE and Aladin only concerns the time-discretization of the adiabatic part of the model so far. These new iterative time-discretizations offer a very wide spectrum of possibilities for the inclusion of the parameterized terms of the evolution inside the time-step, especially when combined to the possibility of fractional vs. sequential approaches.

IFS and ARPEGE/Aladin have made quite different choices for this inclusion of diabatic terms inside the time-step, and it has been decided that harmonization should not concern this part. As a consequence, the part of the shared (between IFS/ARPEGE/Aladin) code concerning the inclusion of diabatic terms into the time-step should stay highly protected by the LEPHYS and LMPHYS switches for time being. For instance, IFS can already use the iterated scheme to perform a more sophisticated inclusion of diabatic terms into the time-step. This generates an extra call of  CALL_SL / LAPINEB  from  GP_MODEL , and allocation of additional arrays. These features should then be protected by the switch LEPHYS. From its side, ARPEGE/Aladin has not yet addressed too much this problem, and a simple algorithm is still applied (the physics is called only once at the predictor iteration, then applied unchanged to all subsequent iterations).

Hence, to summarize, this harmonization of adiabatic time-discretizations is decided not to be the right time to merge or harmonize the inclusion of diabatic terms inside the time-step or to add new functionalities for them. This should be considered as another subject.

Home