the air-sea interaction problems. There may be other approaches, aside from a full high-resolution model, we should be looking at.

MCWILLIAMS: Surface material-property distributions near the Gulf Stream tend to be a pretty broad envelope, since there are both advection by the narrow stream and buffeting by eddies. For climate you need to get the current's transport right, but not necessarily its narrowness and speed maximum.

CANE: I think you'd need to get the advective transport of warmer waters about right in order to get the fluxes right. I think it could be done without introducing tremendous resolution everywhere.

MCWILLIAMS: It's a burden on the sub-grid-scale parameterization. Progress will be made in both resolution and parameterizations, but I think we need to make what you might call the mesoscale calculations before we look at the details of, say, topography. Someone needs to examine topographic parameterization. In the end, of course, it might have to be done with local high-resolution grids for particularly complex regions.

BERGMAN: Is it possible to use variable grids to resolve features like the western boundary currents, yet still have a model that is not excessively complex overall?

BRYAN: Non-uniform grids are used extensively in engineering calculations. But you must then use a very implicit type of calculation, or the time step will be limited by the smallest grid size.

BERGMAN: Are semi-lagrangian approaches being tried for ocean GCMs as well as atmospheric GCMs?

MCWILLIAMS: Yes, some people are pursuing this idea; semi-lagrangian schemes that are sort of shape-preserving keep you from being embarrassed by things like negative concentrations. I myself feel that we can develop multi-grid, fully implicit problem-solvers that can be run at Courant-Friedrichs-Levy numbers of 10 or so, which is a considerable potential economy over semi-lagrangian schemes.

DICKSON: It seems to me that you need a comprehensive observing system to give you a long-period look at the ocean against which you could match a comprehensive climate model's results.

MCWILLIAMS: I quite agree. But please invite the modelers to join the design process!