| 5 | It would be possible to implement different parameterizations and compute them at the same time. |
| 6 | |
| 7 | Roughly the algorithm would be: |
| 8 | |
| 9 | * storage of the state |
| 10 | |
| 11 | * Beginning of loop |
| 12 | |
| 13 | * if different from first loop : restore the state |
| 14 | |
| 15 | * if loop n compute nth model tendencies (the last model must be the model used for actual time stepping) |
| 16 | |
| 17 | * output the nth model variables values to file |
| 18 | |
| 19 | * end of loop |
| 20 | |
| 21 | The storing and restoring is already implemented in the Schwarz loop. |
| 22 | Clearly the Schwarz loop can be modified to do this. |
| 23 | |
| 24 | There are several problems in the case of coupling ! |
| 25 | |
| 26 | You will need to store the fields sent and received instead of going through OASIS for exchanging between ocean and atmosphere. |
| 27 | |
| 28 | If you have several physical time steps during a Schwarz window you will compute a different trajectory for each parameterization instead of a single point. |
| 29 | |
| 30 | In the case of forced LMDZ you can set the Schwarz loop to be for a single physical time step. |
| 31 | |
| 32 | Another solution is to store the ocean forcing of a reference coupling simulation and use this to run a forced LMDZ run. |
| 33 | |