Data such as that collected during the 1991 Convection and Precipitation/Electrification (CAPE) program will be among the most complete for model sensitivity testing and evaluation. Some present views of coastal mesoscale processes may require revision in light of very fine mesh simulations. The issue of just how fine is fine enough remains open.

Four-dimensional data assimilation (4DDA) issues also need to be addressed. While 4DDA has been used with success at meso-alpha and regional scales, the use of data in coastal zones, where extreme vertical and horizontal gradients are common, raises difficult issues of spatial and temporal representativeness of the data to be used for 4DDA. Pielke et al. (1989) demonstrate the need for increased accuracy in measured data as their spatial separation decreases, showing a general inability to relocate time-space cross sections in the coastal environment. Any new observational system installed in coastal zones, if it is to be useful for potential 4DDA applications, must be carefully designed. If operational coastal models are to become a reality, frequently updated regional soil moisture and vegetation information (from observations not now readily available) will become a requirement. Four-dimensional data assimilation techniques currently being investigated include those typified by Errico and Vukicevic (1991), Hoke and Anthes (1976), and Lipton and Pielke (1986).

The following specific recommendations are made:

• Site-specific three-dimensional numerical modeling of coastal zones and studies of coastal zone physics may be enhanced by on-site use of high-performance workstations. The development of massively parallel workstations and user-friendly software to increase this performance capability should be encouraged.

• Standardized procedures to construct subroutines should be developed in order to facilitate the exchange of model components between investigators.

• The value of ensemble model predictions, as contrasted with a collection of model realizations, needs to be determined. Such evaluations should include the testing of subgrid-scale parameterizations that incorporate a stochastic component in order to ascertain the significance of deviations from an ensemble representation.

• The extent to which predictability is extended as a result of the spatial structure of the surface forcing needs to be established.

• Requirements to adequately utilize 4DDA techniques in the coastal environment need to be assessed. Adjoint methods, nudging procedures, and variants of normal model analyses should be tested in the coastal environment where differences in land and water forcing could perhaps be applied to improve the accuracy of data assimilation.