barotropic-baroclinic coupling
With RK3, ORCA2 was unable to run with 3h time step because of instabilities (in particular around the Aleutian Islands). Thanks to Jérôme Chanut, it seems that the issue comes from the coupling between barotropic and baroclinic velocities. Jérôme found that the solution is to avoid interpolation of the barotropic velocities at stages 1 and 2. Hence, replacing those lines in stprk3_stg.F90 (for stage 1 but stage 2 is similar):
uu_b(:,:,Kaa) = r2_3 * uu_b(:,:,Kbb) + r1_3 * ua_b(:,:)
vv_b(:,:,Kaa) = r2_3 * vv_b(:,:,Kbb) + r1_3 * va_b(:,:)
by these ones
uu_b(:,:,Kaa) = ua_b(:,:)
vv_b(:,:,Kaa) = va_b(:,:)
There is also a bug when ln_dynadv_OFF=T, but I am not sure it will be corrected in this branch (Jérôme?)