A primary goal of the Project for Intercomparison of Land-surface Parameterization Schemes (PILPS) [Henderson-Sellers et al., 1996] is to compare land surface schemes used in general circulation models. The first step was to evaluate the different schemes forced by the same atmospheric conditions (Phases 1 and 2). This type of inter-comparison does not take into account the atmospheric feedback which might reduce or increase the differences between schemes. In order to study the possible importance of this feedback all schemes have to be coupled to the same atmospheric model. This is the aim of phase 4 of PILPS, with phase 4a focussed towards general circulation models and 4b focussed towards regional numerical weather prediction models.
At this point in time, coupling a range of land surface schemes to one general circulation model is very labor intensive, due to two main problems. The first one is practical and is linked to the way the FORTRAN code is organized and how variables are managed by individual schemes. The second problem is more fundamental and is related to the fact that the different numerical schemes used to solve the surface energy balance equation require different atmospheric forcing at different time-steps. This difficulty could be overcome by imposing one numerical scheme and asking, for instance, the land surface scheme to provide the land surface fluxes or the parameters needed to compute them (e.g., the ratio between potential and actual evaporation, surface roughness, albedo). This solution raises an important problem, as changes to the numerical framework of the land surface scheme may have a strong impact on the behavior of the scheme. In all surface parameterizations the equations used have been formulated according to the numerical method chosen, or the numerics have been adapted to the equations. Imposing a single numerical framework could destroy this equilibrium and might require a major rewriting of the scheme and in some cases a reformulation of the algorithms employed. The turbulent mixing in the planetary boundary layer is the parameterization where this problem is most acute because its' main forcing is from the surface, the time-scales are very short and the processes are highly non-linear.
This calls for the ``plug compatibility'' of land surface schemes. Kalnay et al. (1989) discussed this issue for the physical parameterizations of general circulation models but in the present case these recommendations need to be extended to a complex subsystem. A general coupling interface for land-surface schemes will be different from the one used for oceans models in OASIS [Terray, 1994] as the time-scales are much shorter thus requiring surface processes to be solved within the time-stepping of the GCM. The flux coupler [Bryan et al., 1997] used at NCAR is another attempt to have one interface to all surfaces. It imposes a numerical scheme which might not be the best choice for in all cases, a severe limitations.
In the present note we would like to describe a method for easily coupling any land surface scheme (LSS) to a general circulation model (GCM) that can accommodate most (if not all) numerical schemes used for modeling surface processes. The first problem which has to be dealt with is to define clearly the tasks of a land surface scheme in order to choose the information which needs to be exchanged between the LSS and the GCM. This will be discussed in the first section. Then we will address the problems raised by coupling radiation and the vertical diffusion in a manner which is independent of the numerical schemes chosen in the GCM and the LSS. Finally it will be described where in the GCM this interface needs to be placed and which other steps can be taken to simplify the coupling.
%