Indexing Values in Continuous Field Databases

• Authors: Myoung-Ah Kang;Christos Faloutsos;Robert Laurini;Sylvie Servigne
• Venue: EDBT '02 Proceedings of the 8th International Conference on Extending Database Technology: Advances in Database Technology
• Year: 2002

With the extension of spatial database applications, during the last years continuous field databases emerge as an important research issue in order to deal with continuous natural phenomena during the last years. A field can be represented by a set of cells containing some explicit measured sample points and by arbitrary interpolation methods used to derive implicit values on nonsampled positions. The form of cells depends on the specific data model in an application. In this paper, we present an efficient indexing method on the value domain in a large field database for field value queries (e.g. finding regions where the temperature is between 20 degrees and 30 degrees). The main idea is to divide a field into subfields [15] in order that all of explicit and implicit values inside a subfield are similar each other on the value domain. Then the intervals of the value domain of subfields can be indexed using traditional spatial access methods, like R*-tree [1]. We propose an efficient and effective algorithm for constructing subfields. This is done by using the field property that values close spatially in a field are likely to be closer together. In more details, we linearize cells in order of the Hilbert value of the center position of cells. Then we form subfields by grouping sequentially cells by means of the cost function proposed in this paper, which tries to minimize the probability that subfields will be accessed by a value query. We implemented our method and carried out experiments on real and synthetic data. The results of experiments show that our method dramatically improves query processing time of field value queries compared to linear scanning.