Grid computing of spatial statistics: Using the teragrid for G*_i (d) analysis

The massive quantities of geographic information that are collected by modern sensing technologies are difficult to use and understand without data reduction methods that summarize distributions and report salient trends. Statistical analyses, therefore, are increasingly being used to analyze large geographic data sets over a broad spectrum of spatial and temporal scales. Computational Grids coordinate the use of distributed computational resources to form a large virtual supercomputer that can be applied to solve computationally intensive problems in science, engineering, and commerce. This paper presents a solution to computing a spatial statistic, *_i(d) using Grids. Our approach is based on a quadtree-based domain decomposition that uses task-scheduling algorithms based on GridShell and Condor. Computational experiments carried out on the TeraGrid were designed to evaluate the performance of solution processes. The Grid-based approach to computing values for G*_i(d) shows improved performance over the sequential algorithm while also solving larger problem sizes. The solution demonstrated not only advances knowledge about the application of the Grid in spatial statistics applications but also provides insights into the design of Grid middleware for other computationally intensive applications.