Spatial Aggregation: Data Model and Implementation

Data aggregation in Geographic Information Systems (GIS) is a desirable feature, only marginally present in commercial systems nowadays, mostly through ad-hoc solutions. Moreover, little attention has been given to the problem of integrating GIS and OLAP (On Line Analytical Processing) applications. In this paper, we first present a formal model for representing spatial data. This model integrates in a natural way geographic data and information contained in data warehouses external to the GIS. This novel approach allows both aggregation of geometric components and aggregation of measures associated to those components, defined in GIS fact tables. We define the notion of geometric aggregation, a general framework for aggregate queries in a GIS setting. Although general enough for expressing a wide range of queries, some of these queries can be hard to compute in a real-world GIS environment. Thus, we identify the class of summable queries, which can be efficiently evaluated by precomputing the overlay of two or more of the thematic layers involved in the query. We also sketch a language, denoted GISOLAP-QL, for expressing queries that involve GIS and OLAP features. In addition, we introduce Piet, an implementation of our proposal, that makes use of overlay precomputation for answering spatial queries (aggregate or not). Piet supports four kinds of queries: standard GIS queries, standard OLAP queries, geometric aggregation queries (like “total population in states with more than three airports”), and integrated GIS-OLAP queries (“total sales by product in cities crossed by a river”, with the possibility of further navigating the results). Our experimental evaluation, discussed in the paper, showed that for a certain class of geometric queries with or without aggregation, overlay precomputation outperforms R-tree-based techniques. This suggests that overlay precomputation can be an alternative to be considered