Completion of OBC BDY merge (MetOffice-1, MetOffice-2 and MERCATOR-6)

[davestorkey]

Work required to complete the merge of the OBC and BDY open boundary modules. Main task: inclusion of Orlanski radiation condition as an option in BDY.

[davestorkey]

The changes are in the following branch:

https://forge.ipsl.jussieu.fr/nemo/browser/branches/2013/dev_r3987_METO1_MERCATOR6_OBC_BDY_merge

Changes included so far:

1. Implementation of Orlanski radiation condition for barotropic velocities, baroclinic velocities and active tracers.
2. Boundary algorithms specified by strings rather than numbers in the namelist to make it more transparent.
3. Introduction of logicals to say whether external data required. (Eg. no-normal gradient condition does not require external data). Simplifies logic in bdy_dta.

Brief notes on Orlanski implementation

---

The implemented version in BDY is from Marchesiello, McWilliams? and Shchpetkin, Ocean Modelling (2001). This has some differences from the OBC version, notably it uses a time-implicit scheme to solve the wave equation on the boundary instead of the time-explicit scheme in OBC. This simplifies the coding since extra prognostic arrays for the boundary variables are not required, and it seems to be more stable.

The Orlanski algorithm is implemented as generic routines in a new bdylib.F90 module. These routines are called from bdy_dyn2d, bdy_dyn3d and bdy_tra to apply the radiation condition on barotropic velocities, baroclinic velocities and tracers respectively. Note that unlike in the OBC version this means that radiation velocities are calculated for each variable separately.

Both the full version of the scheme ("fully oblique radiation") and the approximate version of the scheme ("normal projection of oblique radiation" or NPO) have been coded and can be selected in the namelist as "orlanski" and "orlanski_npo".

As for the OBC version, the scheme has adaptive relaxation to external data in addition to the radiative term, with weak relaxation plus radiation where the radiation velocity is diagnosed outwards and strong relaxation and no radiation where the radiation velocity is diagnosed inwards. The relaxation timescales (in days) for the inward and outward cases are specified in the namelist as rn_time_dmp and rn_time_dmp_out respectively.

So far the Orlanski scheme for the barotropic variables (with adaptive relaxation turned off) has been tested for the equatorial Rossby soliton case detailed in Marchesiello et al. Tests have been done for the linear free surface with both the filtered and time-splitting options. The BDY version of the scheme performs as well as the OBC version where the soliton interacts with the western boundary, and the BDY version seems to be better than the OBC version where kelvin waves interact with the eastern boundary. The BDY version is also more generally stable than the OBC version. I plan to present the results in a short report.

Further testing will be done for the non-linear free surface (VVL) and full 3D models.