An important component in the calculation of the turbulent surface fluxes is the diffusion coefficient, which depends on the surface roughness and on the stability of the lower layers of the atmosphere. The first component is within the domain of the land-surface scheme but varies only at a seasonal scale, while the second one is more atmospheric in nature and evolves rapidly.
Some LSS compute the diffusion coefficient, while in others the value is obtained from the vertical diffusion parameterization of the GCM. Both methods have their advantages which justifies that the interface should not exclude either one. On the other hand each one has problems which might affect the coupled system and should be controlled by the interface.
Because of the continuity of the turbulent surface fluxes and the vertical diffusion within the planetary boundary layer (PBL), allowing the LSS to compute its own diffusion coefficient risks losing the consistency of the diffusion coefficients. The parameterizations of the planetary boundary layer used in GCMs may be different from the one of the LSS. Thus if the LSS is allowed to compute its own diffusion coefficient, the mismatch with the GCM might affect the result. For instance, moisture or heat might accumulate in an unrealistic way at the lowest level only because the formulation of the surface layer produces a diffusion much too large for the PBL. But LSSs which solve the surface energy balance through iterations need to recompute the stability function included in the formulation of the surface transfer coefficient to improve convergence. This can only be done if the LSS has control over this formulation.
When the surface layer diffusion coefficients are computed by the GCM, modeling the heterogeneities at the surface is limited. The PBL scheme of the GCM only provides one diffusion coefficient per grid-box and more subtle methods are needed to propagate the surface heterogeneities into the surface layer and the PBL.
At present it is not possible to decide if one approach should be favored over the other and it is reasonable to allow for both solutions, but work is needed on this issue.