Organization of data in non-convex spatial domains

Abstract

We present a technique for organizing data in spatial databases with non-convex domains based on an automatic characterization using the medial-axis transform (MAT). We define a tree based on the MAT and enumerate its branches to partition space and define a linear order on the partitions. This ordering clusters data in a manner that respects the complex shape of the domain. The ordering has the property that all data down any branch of the medial axis, regardless of the geometry of the sub-region, are contiguous on disk. Using this data organization technique, we build a system to provide efficient data discovery and analysis of the observational and model data sets of the Chesapeake Bay Environmental Observatory (CBEO). On typical CBEO workloads in which scientists query contiguous substructures of the Chesapeake Bay, we improve query processing performance by a factor of two when compared with orderings derived from space filling curves. © 2010 Springer-Verlag Berlin Heidelberg.