Tests of tracer conservation in NEMO (V3.4 Off line mode, MY_TRC ) failed. I have no idea of the origin of the problem, whose magnitude highly depends on tracer gradients and fluxes.

I performed the tests with 6 tracers differing by their initial state; further 2 of them undergo exchange at the air-sea interface. In the latter case the gross flux is constrained to be nul. Nevertheless the drift is significant for all these tracers, but the one initialized with a constant value (tracer 1).

It seems advection is a possible culprit. I tested several schemes, none lead to satisfactory results.

I attach the file containing details on how i proceeded as well as the drift results.

In closea.F90 / clo_rnf

p_rnfmsk experiences out of bounds index. mi0 and mj0 should not be used directly, but only to define a range of index.

Here is my proposal of clo_rnf :

  SUBROUTINE clo_rnf( p_rnfmsk )
      !!---------------------------------------------------------------------
      !!                  ***  ROUTINE sbc_rnf  ***
      !!                    
      !! ** Purpose :   allow the treatment of closed sea outflow grid-points
      !!                to be the same as river mouth grid-points
      !!
      !! ** Method  :   set to 1 the runoff mask (mskrnf, see sbcrnf module)
      !!                at the closed sea outflow grid-point.
      !!
      !! ** Action  :   update (p_)mskrnf (set 1 at closed sea outflow)
      !!----------------------------------------------------------------------
      REAL(wp), DIMENSION(jpi,jpj), INTENT(inout) ::   p_rnfmsk   ! river runoff mask (rnfmsk array)
      !
      INTEGER  ::   jc, jn, ji, jj      ! dummy loop indices
      !!----------------------------------------------------------------------
      !
      DO jc = 1, jpncs
         IF( ncstt(jc) >= 1 ) THEN            ! runoff mask set to 1 at closed sea outflows
             DO jn = 1, 4
                DO jj =    mj0( ncsjr(jc,jn) ), mj1( ncsjr(jc,jn) )
                   DO ji = mi0( ncsir(jc,jn) ), mi1( ncsir(jc,jn) )
                      p_rnfmsk(ji,jj) = MAX( p_rnfmsk(ji,jj), 1.0_wp )
                   END DO
                END DO
            END DO 
         ENDIF 
      END DO 
      !
   END SUBROUTINE clo_rnf

It seems that agrif_user.F90 routine has not been updated during the merging process (modif made for compatibility with split-explicit free surface).