dynadv_ubs.F90 : problem with lateral boundary condition in version 3.6 (with bugfix)

[ducousso]

• idealized experiments based on one of the COMODO test-case suggest that the current treatment of lateral boundary condition within the UBS momentum advection scheme is doubtful.

• under free-slip condition, the evolution of a vortex impinging a side-wall departs qualitatively from the expected behavior.

• the joint movie cmp_evol_C4_vs_current_UBS.mpeg illustrates the problem with two identical experiments except for the momentum advection-diffusion schemes.

• the left pannel uses the centered 4th order scheme (gamma1 = 0._wp and gamma2 = 1._wp/32._wp in dynadv_ubs.F90) for advection and the biharmonic scheme for diffusion (rn_ahm_0_blp = -2.5e8)) while the right pannel uses the UBS scheme (gamma1 = 1._wp/3._wp and gamma2 = 1._wp/32._wp in dynadv_ubs.F90)) for both advection and diffusion ; in both cases, the boundary condition is free-slip (rn_shlat = 0.)).

• Laurent Debreu (LJK, Grenoble) has suggested the following modification of dynadv_ubs.F90.

change

zlu_uu(ji,jj,jk,1) = ( ub (ji+1,jj,jk)-2.*ub (ji,jj,jk)+ub (ji-1,jj,jk) ) * umask(ji,jj,jk)
zlv_vv(ji,jj,jk,1) = ( vb (ji,jj+1,jk)-2.*vb (ji,jj,jk)+vb (ji,jj-1,jk) ) * vmask(ji,jj,jk)
zlu_uv(ji,jj,jk,1) = ( ub (ji,jj+1,jk)-2.*ub (ji,jj,jk)+ub (ji,jj-1,jk) ) * umask(ji,jj,jk)
zlv_vu(ji,jj,jk,1) = ( vb (ji+1,jj,jk)-2.*vb (ji,jj,jk)+vb (ji-1,jj,jk) ) * vmask(ji,jj,jk)
!
zlu_uu(ji,jj,jk,2) = ( zfu(ji+1,jj,jk)-2.*zfu(ji,jj,jk)+zfu(ji-1,jj,jk) ) * umask(ji,jj,jk)
zlv_vv(ji,jj,jk,2) = ( zfv(ji,jj+1,jk)-2.*zfv(ji,jj,jk)+zfv(ji,jj-1,jk) ) * vmask(ji,jj,jk)
zlu_uv(ji,jj,jk,2) = ( zfu(ji,jj+1,jk)-2.*zfu(ji,jj,jk)+zfu(ji,jj-1,jk) ) * umask(ji,jj,jk)
zlv_vu(ji,jj,jk,2) = ( zfv(ji+1,jj,jk)-2.*zfv(ji,jj,jk)+zfv(ji-1,jj,jk) ) * vmask(ji,jj,jk)

to

zlu_uu(ji,jj,jk,1) = ( ub (ji+1,jj,jk)-2.*ub (ji,jj,jk)+ub (ji-1,jj,jk) ) * umask(ji,jj,jk)
zlv_vv(ji,jj,jk,1) = ( vb (ji,jj+1,jk)-2.*vb (ji,jj,jk)+vb (ji,jj-1,jk) ) * vmask(ji,jj,jk)
zlu_uv(ji,jj,jk,1) = ( ub (ji,jj+1,jk)-ub (ji,jj,jk) ) * fmask(ji,jj,jk)-( ub (ji,jj,jk)-ub (ji,jj-1,jk) ) * fmask(ji,jj-1,jk)
zlv_vu(ji,jj,jk,1) = ( vb (ji+1,jj,jk)-vb (ji,jj,jk) ) * fmask(ji,jj,jk)-( vb (ji,jj,jk)-vb (ji-1,jj,jk) ) * fmask(ji-1,jj,jk)
!
zlu_uu(ji,jj,jk,2) = ( zfu(ji+1,jj,jk)-2.*zfu(ji,jj,jk)+zfu(ji-1,jj,jk) ) * umask(ji,jj,jk)
zlv_vv(ji,jj,jk,2) = ( zfv(ji,jj+1,jk)-2.*zfv(ji,jj,jk)+zfv(ji,jj-1,jk) ) * vmask(ji,jj,jk)
zlu_uv(ji,jj,jk,2) = ( zfu (ji,jj+1,jk)-zfu (ji,jj,jk) ) * fmask(ji,jj,jk)-( zfu (ji,jj,jk)-zfu (ji,jj-1,jk) ) * fmask(ji,jj-1,jk)
zlv_vu(ji,jj,jk,2) = ( zfv (ji+1,jj,jk)-zfv (ji,jj,jk) ) * fmask(ji,jj,jk)-( zfv (ji,jj,jk)-zfv (ji-1,jj,jk) ) * fmask(ji-1,jj,jk)

• the bugfix allows to recover a free-slip character for UBS scheme close to the boundary as illustrated in the joint movie cmp_evol_C4_vs_modified_UBS.mpeg.

[smasson]

change done in changeset:5061