Might have found small bug in the vertical tracer gradient used in the horizontal redi diffusivity

[patrickscholz]

I think to compute the vertical tracer gradient we need here hnode and not hnode_new. hnode_new at this point already contains the new vertical grid (in case of zstar) but i think the horizontal redi diffusivity is still done on the old vertical grid. Since we also use helem (old vertical grid ) to compute the horizontal diffusive fluxes.

Am i right? In this case it would change our default testcase!

This problem only account for the zstar and zlevel case

fesom2/src/oce_tracer_mod.F90

Lines 118 to 130 in 2695a5d

	DO n=1, myDim_nod2D+eDim_nod2D
	!!PS nlev=nlevels_nod2D(n)
	nzmax=nlevels_nod2D(n)
	nzmin=ulevels_nod2D(n)
	!!PS DO nz=2, nlev-1
	DO nz=nzmin+1, nzmax-1
	dz=0.5_WP*(hnode_new(nz-1,n)+hnode_new(nz,n))
	tr_z(nz, n)=(ttf(nz-1,n)-ttf(nz,n))/dz
	END DO
	!!PS tr_z(1, n)=0.0_WP
	!!PS tr_z(nlev, n)=0.0_WP
	tr_z(nzmin, n)=0.0_WP
	tr_z(nzmax, n)=0.0_WP

[patrickscholz]

just spoke with Sergey about, so its definitely a bug and needs to be changed. Not sure how large is its implication, most likely not to big.

[patrickscholz]

I think i just might have found another issue regarding Redi diffusivity ...

fesom2/src/oce_ale_tracer.F90

Lines 960 to 1000 in 2695a5d

	!$OMP END DO
	! no halo exchange of tr_xynodes is needed !
	!$OMP DO
	do n=1, myDim_nod2D
	nl1=nlevels_nod2D(n)-1
	ul1=ulevels_nod2D(n)
	vd_flux=0._WP
	!_______________________________________________________________________
	zbar_n(1:mesh%nl )=0.0_WP
	z_n (1:mesh%nl-1)=0.0_WP
	zbar_n(nl1+1)=zbar_n_bot(n)
	z_n(nl1)=zbar_n(nl1+1) + hnode_new(nl1,n)/2.0_WP
	do nz=nl1, ul1+1, -1
	zbar_n(nz) = zbar_n(nz+1) + hnode_new(nz,n)
	z_n(nz-1) = zbar_n(nz) + hnode_new(nz-1,n)/2.0_WP
	end do
	zbar_n(ul1) = zbar_n(ul1+1) + hnode_new(ul1,n)
	!_______________________________________________________________________
	do nz=ul1+1,nl1
	vd_flux(nz)=(z_n(nz-1)-zbar_n(nz))*(slope_tapered(1,nz-1,n)*tr_xynodes(1,nz-1,n)+slope_tapered(2,nz-1,n)*tr_xynodes(2,nz-1,n))*Ki(nz-1,n)
	vd_flux(nz)=vd_flux(nz)+&
	(zbar_n(nz)-z_n(nz)) *(slope_tapered(1,nz,n) *tr_xynodes(1,nz,n) +slope_tapered(2,nz,n) *tr_xynodes(2,nz,n)) *Ki(nz,n)
	vd_flux(nz)=vd_flux(nz)/(z_n(nz-1)-z_n(nz))*area(nz,n)
	enddo
	do nz=ul1,nl1
	del_ttf(nz,n) = del_ttf(nz,n)+(vd_flux(nz) - vd_flux(nz+1))*dt/areasvol(nz,n)
	enddo
	end do
	!$OMP END DO
	!$OMP END PARALLEL
	end subroutine diff_ver_part_redi_expl
	!
	!
	!===============================================================================
	subroutine diff_part_hor_redi(tracers, partit, mesh)
	use o_ARRAYS
	use MOD_MESH
	USE MOD_PARTIT

... also here when computing zbar_n and z_n we must use the old hnode not the new one, since its the explicte part of Redi ?!

[christian-stepanek]

Hi @patrickscholz, is it known whether or not the bug leads to problems so severe that one should upgrade the FESOM2 source code asap? If it is severe, is there already an accepted fix? Thanks a lot for reporting this!!!

[patrickscholz]

@christian-stepanek, I havent tested yet the full severity of that issue since i had other things to do before the holidays. I just stumbelt over these things while working on another problem. Maybe i manage to make some test runs in the next weeks! The problem here is also that this will change our fesom testcases for binary identiy. When we merge this these need to be done new

[christian-stepanek]

Thanks a lot @patrickscholz, I will keep my eyes open re any updates on this issue.