should involve fields beyond simple velocity fields or buoy tracks for comparison;for example, the convergence/divergence field presents a difficult area for a model to capture, and yet it has been observed without too much difficulty in the field.
Finally, convection received little mention in the first day's talks, and yet forms a backbone of much of the shelf-slope exchange. More study remains necessary here. The mechanism supporting the halocline came in for much discussion: it was noted that two basins existed, giving a useful test for any process model. (Indeed, the lack of horizontal homogeneity of the Arctic basin implied that different dynamics must occur in different regions.) A hierarchy of models is needed, with a focus on a subset of issues. Model output is frequently sensitive to spatial gradients, so that sensitivity studies are vital.
Observations
The question of how best to observe a process remains thorny. Too often, meetings have generated some apparently optimal combination of instruments or platforms, without attention to whether adequate resources existed to tackle the scientific question in the first place. (The meeting did not attempt to define a strategy, which was probably a wise choice.) Nonetheless, we should address whether sections, leading to two-dimensional thinking for what is definitely a three-dimensional problem, or limited area surveys, which contain their own distortions precisely because they are limited in both space and time, are preferable. Three-dimensional sampling may suffer from short coherence scales. Neither two- nor three-dimensional sampling may be perfect for estimating integrated quantities such as transports, which are known in other areas to be concentrated in narrow areas and are often missed by mooring designs. In a similar fashion, sudden events in time can equally easily be missed unless data are taken almost continuously. What, furthermore, are the best observations to validate or falsify a model (even a conceptual model) of shelf-slope interactions?
Issues
Much attention was given to how water masses leave the shelf region; yet this volume of water must be replaced somehow, not necessarily always by simple advection from a distant ocean. Preconditioning was mentioned often as a necessary part of the shelf-slope convection process. We remain unclear what form this takes. The need to distinguish trends from long-term fluctuations was emphasised. In particular, in a red spectrum, what exactly is meant by a "mean"? Can we make use of conserved quantities to diagnose the circulation? (Potential vorticity, as well as passive tracers, come to mind here.)
Finally some modelling issues: How should we parameterise shelf effects for climate models? Is the best starting point a coastal or a deep model, or a new combination
