|
Next: 2. Harmonization of iterative Up: Main page Previous: Main page 1. IntroductionIn addition to the classical SI time-discretization, several iterative time-discretizations have been developed more or less independently in IFS and in ARPEGE/Aladin. Although they are not to be used operationally in the immediate future, they represent an investment for future potential use (not sure in next years for IFS, almost sure for Aladin-NH). This is because better accuracy and/or stability properties are expected from them.
These alternative schemes are as follows: developed in IFS: - LFULLIMP developed in Aladin: - LPC_OLD - LPC_FULL - LPC_TRAJ
The scheme LPC_OLD is the so-called partial iterative scheme which has been use for Aladin-NH since the beginning of Aladin-NH implementation (Bubnova et al, 1995). The schemes LFULLIMP and LPC_TRAJ are based on the same concept of an iterative approximation of the trapezoidal scheme, including the calculation of SL trajectories. However, LFULLIMP and LPC_TRAJ even if exploiting the same basic idea, are written (and coded) quite differently: LFULLIMP is written in a specific "incremental form", while LPC_TRAJ is written in the same "non-incremental form" as the traditional SI scheme of IFS/ARPEGE/Aladin (see Appendix A). The LPC_FULL is also an iterative approximation of a trapezoidal scheme, but the trajectories are computed only from the explicit winds, as for the SI scheme. This scheme is less sophisticated (or consistent) because trajectories are not re-computed at each iteration, but it is less expensive, and could be advantageous in the cases where the trajectory iteration is not the main concern.
The new current implementations of these schemes are actually coded in "research" form not fully rationalized, and not fully integrated in a clean way into the main stream of cycles. This means in particular that:
Since LPC_FULL and LPC_TRAJ schemes are similar, they should be "harmonized", i.e. merged as deeply as possible. Moreover, there is clearly a need of a better integration of these schemes in the main code. However, this requires an effort of rationalization for smooth acceptance in the official cycles by the other persons in charge of the code. This note lists the points which have been thought as necessary to achieve this objective.
Next: 2. Harmonization of iterative Up: Main page Previous: Main page Pierre BENARD 2002-06-17
|