Modelling of runoff in GCMs is a two-stage process: first, a residual water amount resulting from precipitation, evapotranspiration, snowmelt, infiltration and soil moisture movement at each time step must be generated for each grid cell; then a runoff routing model must be applied to this water excess to produce streamflow. Incorporation of the latter into GCMs is as yet in the most initial stages, and will not be reliably achieved for some years to come. However, before this can be accomplished, the water excess that is to be routed must be generated in a physically realistic fashion.
Two land-surface models of widely varying complexity are presently available as options in the Canadian GCM (developed at Atmospheric Environment Service, Canada). The old scheme is based on a bucket approach for soil moisture and on the force-restore method for soil temperature; a more recently developed scheme (“CLASS” – Canadian Land Surface Scheme) incorporates physical modelling of the heat and moisture budgets of three soil layers, the snow cover and the vegetation canopy. Based on comparisons of year-long runs of both models, it is postulated that reliable modelling of the surface water excess cannot be achieved using a “bucket” model. It is also shown that sub-grid scale variability of the precipitation rate must be taken into account, and that the land surface model must allow for surface water retention between timesteps.