misplaced modification of fstravs in trczdf_iso.F90

[sga]

In routine trczdf_iso.F90 the vertical part of the isopycnal mixing is added to the fstravs array inside a loop over the number of tracers, before fstravs is used in a second loop over the tracers. This means that if there are 10 tracers, that the isopycnal mixing is 10 times too big. It needs to be moved outside the loop over jn. A suggested patch against revision 1793 of the trunk (identical to our revision 1109) would be:

Index: NEMO/TOP_SRC/TRP/trczdf_iso.F90
===================================================================
--- NEMO/TOP_SRC/TRP/trczdf_iso.F90 (revision 1109)
+++ NEMO/TOP_SRC/TRP/trczdf_iso.F90 (working copy)
@@ -180,8 +180,22 @@

!---------------------------------------------------
zavs(:,:,:) = fstravs(:,:,:)

+ ! 0.2 update and save of avt (and avs if double diffusive mixing)
+ ! ---------------------------

+ DO jk = 2, jpkm1
+ DO ji = 2, jpim1

+ zavi = fsahtw(ji,jj,jk)*( wslpi(ji,jj,jk)*wslpi(ji,jj,jk) &
+ & +wslpj(ji,jj,jk)*wslpj(ji,jj,jk) )
+
+ ! add isopycnal vertical coeff. to avs
+ fstravs(ji,jj,jk) = fstravs(ji,jj,jk) + zavi
+
+ END DO
+ END DO
+
+

DO jn = 1, jptra

IF( l_trdtrc ) ztrtrd(:,:,:) = tra(:,:,:,jn) ! save trends

@@ -262,21 +276,6 @@

END DO

• ! I.3 update and save of avt (and avs if double diffusive mixing)
• ! ---------------------------

-

• DO jk = 2, jpkm1
• DO ji = 2, jpim1

-

• zavi = fsahtw(ji,jj,jk)*( wslpi(ji,jj,jk)*wslpi(ji,jj,jk) &
• & +wslpj(ji,jj,jk)*wslpj(ji,jj,jk) )

-

• ! add isopycnal vertical coeff. to avs
• fstravs(ji,jj,jk) = fstravs(ji,jj,jk) + zavi

-

• END DO
• END DO

-

#if defined key_trcldf_eiv

! ! ---------------------------------------!
! ! Eddy induced vertical advective fluxes !

[cetlod]

The same problem are in the routine trczdf_iso_vopt.F90 and trczdf_imp.F90. The results are right because temporary arrays used to compute the vertical diffusion coefficient are re-initialize to zero in the loop over the tracers.

We should apply the same patch suggested by Steven to correct all these routines.

Thanks Steven

[acc]

Replying to cetlod:

The same problem are in the routine trczdf_iso_vopt.F90 and trczdf_imp.F90. The results are right because temporary arrays used to compute the vertical diffusion coefficient are re-initialize to zero in the loop over the tracers.

"the results are right"?
It looks to me as if only the edge points of zavsi are reset each iteration so surely trczdf_iso_vopt suffers from the same accumulation error?

[cetlod]

• In trczdf_iso_vopt.F90, we use a temporay array,"zavsi", which contains the vertical mixing coef fstravs + the vertical part of the isopycnal mixing

DO LOOPS

zavi = fsahtw(ji,jj,jk)*( wslpi(ji,jj,jk)*wslpi(ji,jj,jk) &
& +wslpj(ji,jj,jk)*wslpj(ji,jj,jk) )
! add isopycnal vertical coeff. to avs
zavsi(ji,jj,jk) = fstravs(ji,jj,jk) + zavi

ENDDO LOOPS

• In trczdf_iso.F90, we update fstravs with the the vertical part of the isopycnal mixing

DO LOOPS

zavi = fsahtw(ji,jj,jk)*( wslpi(ji,jj,jk)*wslpi(ji,jj,jk) &
& +wslpj(ji,jj,jk)*wslpj(ji,jj,jk) )
! add isopycnal vertical coeff. to avs
fstravs(ji,jj,jk) = fstravs(ji,jj,jk) + zavi

ENDDO LOOPS

That's the real reason why we accumulate in trczdf_iso.F90 and not in trczdf_iso_vopt.F90