This paper deals with data structures within GIS. Continuous phenomenons are usually represented by raster structures for simplicity reasons. With such structures spatial repartitions of the data is not easily interpretable. Moreover, in an overlapping clustering context these structures remove the links between the data and the algorithms. We propose a vector representation of such data based on non-regular multi-rings polygons. The structure requires multipart nested polygons and new set operations. We present the formalism based on belief theory and uncertainty reasoning. We also detail the implementation of the structures and the set operations. The structures and the set operations are illustrated in the context of forest classification having diffuse transitions.