In the last few years some members of the GIS community have argued that the raster/vector debate obscures more essential issues about how people think about and manipulate space. Suppose that you want to describe the agricultural activity in a certain region. There are several spatially-referenced variables that may be of interest: owner, crop grown, soil type, elevation, rainfall, etc.
One might speak of farms in the region as geometric objects that have owner attributes. Fields could be objects with a crop grown attribute, and could inherit an owner from the farms that contain them. This is an example where an object-oriented approach to the data is natural: There is clear separation of objects, their representation can be encapsulated, and they form a hierarchy in which representations and attributes can be inherited.
When one adds soil type, however, one may find that a field can contain several soil types or that a soil type may encompass several fields. Subdividing fields into objects that can be given a single soil type attribute may or may not be a good idea-it depends on the importance of the soil type variable. It seems more natural to start a new layer. This in turn creates problems where the same object may have different representations in different layers. A stream, for example, may be represented as a change of ownership and a change in soil type.
It might be natural to think of soil type, elevation, and rainfall as functions defined over the region-cropland as a ``vector field.'' Defining data structures to support continuous data models is difficult because data is usually obtained by discrete sampling. Even so, the amount of data can be unwieldy and data can be expensive to collect. The traditional soil map quantizes the domain into subjective categories (clay, clay with gravel, sandy, very sandy, etc.) and creates a raster or partitions the region into cells that are assigned one of these soil types to obtain an object-oriented, vector-based map. An implementation of a vector field idea might represent the proportion of each possible component (clay, gravel, sand, etc.) as elements of a vector. One then needs algorithms to quantize this vector space to present understandable output.
Arguments in this debate usually conclude by recommending raster or vector implementation so that the object/layer/field debate, too, becomes limited by current technology.