Phenomena of the real world are described in a variety of forms in current geographic data bases (GDBs) : geographic data models, users' points of view, systems are different. Concurrently re-using GDBs thus requires an integration process both to eliminate duplicates and to regroup complements. Integration makes it possible to federate data from different sources while cutting down acquisition costs (new data captures are avoided) ; it is a crucial issue for interoperability between GDBs.After integration, several representations of real world phenomena are available, with distinct points of view and at different scales. These multiple representations are useful and even necessary for a wide range of applications, such as multi-scale cartography, update propagation, aided navigation etc.The aim of the thesis is to devise an integration process on 2-D, vector data of a single-site GDB. It is modelled as an extended classical three-stepped integration process (schema preparation, investigation for correspondences, integration) [Spaccapietra et al. 92]. The extension involves a taxonomy of integration conflicts between GDBs and a process for joint geometric / topologic data matching. The integration process has been applied from IGN's three main data bases (BD TOPO®, BD CARTO® and GEOROUTE®) on the Lagny area (900 km of roads network).Given the complexity of the real world's phenomena, several versions and as many models may be defined. The differences are structured in the taxonomy of integration conflicts : conflicts come as definition conflicts (classification conflicts, fragmentation conflicts, specification conflicts), heterogeneity conflicts, description conflicts... Altogether, six conflict categories have been used for the integration process.Some conflicts are taken into account at the stage of schema preparation. Other conflicts require specific treatment : extension of the language of correspondence declarations and operations to solve the conflicts. Indeed, integration requires a strategy. The strategy makes the choice of operations to perform and fixes the goal of the integration process. For the databases of the experiments, two strategies and their associated operations are shown.The data matching process consists in identifying sets of data representing the same real world phenomenon and allows to regroup data. This step is precious for it enhances GDBs with inter-representation operations that are useful for multi-representation applications.A data matching process has been developed for road data at different scales, with 90% of the results correct. Henceforth, a generic process has been inferred to help designing matching processes on other kinds of data.The thesis, describing a generic and detailed framework for the integration of GDBs, contributes to the development not only of multi-representation applications but also of interoperability between GDBs , once the processes are adapted to network distributed GDBs.
