Two Algorithms of Searching Spatially Continuous Mark Clusters in Marked Point Process

• Activity Number: 126
• Type: Contributed
• Date/Time: Monday, August 3, 2009 : 8:30 AM to 10:20 AM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Huaguo Wang*+ and Michael Schucking
• Companies: QinetiQ North America/Technology Solutions Group/PSI and QinetiQ North America/Technology Solutions Group/PSI
• Keywords: marked point process ; spatial clusters ; clustering algorithms ; homogeneous subregion ; dissimilarity ; crop insurance

Abstract

For a marked point process M = {[xi, m{xi)]}, where m(xi) is the mark of the point xi, the aim of analysis is to search large spatially continuous mark clusters with small mark variances. In the analysis, xi and m(xi) are not equivalent, hence the typical spatial clustering algorithms based on weighted dissimilarity(ies) of xi and m(xi) cannot be used. Two algorithms, an agglomerative hierarchical algorithm and a local cluster growing algorithm, were developed to conduct such analysis. In each clustering step, both of them use xi to search the neighbor cluster(s) for an existing cluster and use m(xi) to determine whether the neighbor cluster(s) is merged into the existing cluster. Application of the algorithms was demonstrated using multiple-year's crop insurance data, in which the target is to determine spatially homogeneous areas of crop yields in the USA counties.