ln_mskland: special u-, v-, w-mask used ?

[mathiot]

Context

When ln_mskland = .true., the mask used for u-,v- and w-points is not the u-, v-, w-mask. It is a special mask build to let unmask the value along the coastlines as well as the level mbathy+1. So the consequences are that you have fill_value over land, 0 along coastline (except for utau) and ocean value. So, the u-, v-, w-mask rebuild from fill_value will not fit the mask in mesh_mask.nc. So it could lead to wrong diagnostics if fill_value is used (mean value over a box for example)

I found it a bit weird. Why do we apply this special treatment in the output ?

Analysis

Recommendation

If there is no good reason, in iom.F90, I suggest to use u-,v- and w-mask instead of

CASE('U')   ;   zmask(2:jpim1,:,:) = tmask(2:jpim1,:,:) + tmask(3:jpi,:,:)   ;   CALL lbc_lnk( zmask, 'U', 1. )
CASE('V')   ;   zmask(:,2:jpjm1,:) = tmask(:,2:jpjm1,:) + tmask(:,3:jpj,:)   ;   CALL lbc_lnk( zmask, 'V', 1. )
CASE('W')   ;   zmask(:,:,2:jpk  ) = tmask(:,:,1:jpkm1) + tmask(:,:,2:jpk)   ;   zmask(:,:,1) = tmask(:,:,1)

[smasson]

The idea of this special treatment was that coastal u-v values have a physical meaning. For example, there is no current going through the coast, so u=0 for a north-south coastline is a reality and has known value (=0). This is not a undefined value and we decided to keep these coastal values in the outputs. 

Note that this is not the case for coastal utau/vtau that are not used and are undefined. So we masked the coastal value of this field.

[mathiot]

Thank you for the precision.