Design and Implementation of Multi-scale Databases

The need to access spatial data at multiple levels of detail is a fundamental requirement of many applications of geographical information, yet conventional spatial database access methods are based on single resolution spatial objects. In this paper we present a design for multi-scale spatial objects in which both spatial objects and the vertices of their component geometry are labelled with scale priority values. Alternative approaches to database implementation are considered in which vertices are organised into scale-bounded layers. Access times for spatially-indexed vertex block schemes (comparable to the PR-File) were superior to a BLOB scheme where only entire multi-scale objects were spatially indexed. The use of a 3D R-tree to integrate scale and space indexing was found to improve considerably on using either R-Tree indexing of space only or B-tree indexing of scale. Techniques are also presented for client-side reconstruction of cached multi-scale geometry. Implementations are compared in a client-server environment using the Informix object relational database system.