There is a sign error in the lid melting code in the topographic melt pond scheme in SI3. In the current version of NEMO/trunk/src/ICE/icethd_pnd.F90 (last changed at [14997]) the error is in line 664. This currently reads

zdvice = MIN( dh_i_sum_2d(ji,jj,jl)*a_ip(ji,jj,jl), v_il(ji,jj,jl) )

but should be

zdvice = MIN( -dh_i_sum_2d(ji,jj,jl)*a_ip(ji,jj,jl), v_il(ji,jj,jl) )

(compare line 597). In the code as currently written, zdvice is always negative or zero, with the result that the test at line 666 always fails and the code that transfers volume from the lid to the pond is never executed. I've confirmed this by adding a statement that writes a log message the first time this piece of code is executed.

For ISF when using option cn_gammablk=vel_stab, one must use the variable qoce in the subroutine isfcav_gammats but it is only calculated afterwards in isfcav_mlt. So, one must initialize qoce before its use in isfcav_gammats. In addition the initialization of qoce_b is probablement not the right one. It is written as the sensible flux while it should be the latent flux (see isfcav_mlt routine).

Fix: change this line:

      IF (TRIM(cn_gammablk) == 'vel_stab' ) zqoce_b (:,:) = ptsc(:,:,jp_tem) * rho0_rcp ! last time step total heat fluxes (to speed up convergence)

by this one:

      IF ( TRIM(cn_gammablk) == 'vel_stab' ) THEN
         zqoce(:,:) = -pqfwf(:,:) * rLfusisf ! 
         zqh_b(:,:) =  ptsc(:,:,jp_tem) * rho0_rcp ! last time step total heat fluxes (to speed up convergence)
      ENDIF

And make the convergence criterion on qoce+qhc qnd not only qoce:

            zerr = MAXVAL(ABS(zqhc(:,:)+zqoce(:,:) - zqh_b(:,:)))

I'll do the changes and commit them because this routine just does not pass debug right now but it needs to be checked out by people who know (I guess Pierre)