When an overlay is carried out, it is assumed that the comparison is legitimate, spatially and temporally. For example, it would be reasonable to overlay a geology map with a vegetation map. However, if the geology map described the bedrock as it was 200 million years ago and the vegetation map referred to current conditions, the comparison would be meaningless. When we overlay, typically we are looking at the conjunction of different phenomena in both time and space. Alternatively we may have different snapshots in time of the same region of space. The former case pertains to the decision unit - PLU overlay, and consequently, we must assume that the decision units exist at the time the satellite image was taken.
All overlays should not necessarily be assumed to be best represented by multiple inheritance. In the Clayoquot example the overlay involved two different phenomena which both pertained to the same space and the same time. The resultant units, existing in space at a given time, could be viewed as specialized versions of the parent units. Any question that could be posed to each parent could also be posed to the child.
Consider next the watersheds in the Clayoquot. It may be of interest to consider watersheds as hydrologic objects. We can envision such queries as what is the estimated discharge at the outlet of the watershed. We might also consider each watershed as a wildlife habitat object. If we had previously defined a hydrologic class and a habitat class, then Watershed could be considered to be derived from both, i.e., use of multiple inheritance would be appropriate.
In another scenario habitat areas might be defined independently of the watershed boundaries, or the they might involve much more detailed mapping. In either of these cases we may wish to overlay the two coverages. The resultant units however are no longer watersheds - they are parts of watersheds, inappropriate for addressing a number of hydrologic queries. In this case multiple inheritance should not be used. When breaking something up into smaller parts, we must decide whether the semantics of the constituent pieces are the same as for the whole object. In this particular case, pieces of a watershed are different conceptually from an entire watershed.
Multiple inheritance can be a powerful tool with spatial and temporal data as well as many other kinds of information. Nevertheless, considerable care must be used if the results are to be valid. The following overlapping Points are of interest if multiple inheritance is being considered.