An anytime revision operator for large and uncertain geographic data sets

Abstract

 The environmental data are in general imprecise and uncertain, but they are located in space and therefore obey to spatial constraints. The “spatial analysis” is a (natural) reasoning process through which geographers take advantage of these constraints to reduce this uncertainty and to improve their beliefs. Trying to automate this process is a really hard problem. We propose here the design of a revision operator able to perform a spatial analysis in the context of one particular “application profile”: it identifies objects bearing a same variable bound through local constraints. The formal background, on which this operator is built, is a decision algorithm from Reiter [9]; then the heuristics, which help this algorithm to become tractable on a true scale application, are special patterns for clauses and “spatial confinement” of conflicts. This operator is “anytime”, because it uses “samples” and works on small (tractable) blocks, it reaggregates the partial revision results on larger blocks, thus we name it a “hierarchical block revision” operator. Finally we illustrate a particular application: a flooding propagation. Of course this is among possible approaches of “soft-computing” for geographic applications.