This paper presents a knowledge-based system for land use interpretation and prediction. We describe our needs for representing knowledge and data, and for reasoning. We explain our choices : case-based reasoning within the framework of the description logic system RACER. Then, we present the knowledge base and the data we are working with. Data about spatial entities are represented as graphs and represented in the DL system accordingly. An example of graph manipulation is used to illustrate our purpose. Then, we propose a first synthesis of our research work and present an extension of the DL system necessary for going further.