Getis and Ord (1992) introduced a statistic, Gi*(d), that may be used to assess spatial dependence. Currently, we focus on a simple version of the standardized form of this statistic that was further elucidated by Ord and Getis (1995). In a typical exploratory geographic analysis, Gi*(d) is calculated for several different values of d in order to detect spatial clustering at different scales. It is clear that Gi*(d) is a local measure. As such, it is particularly helpful when applied to datasets for which global measures of spatial dependence, such as Moran’s I (Cliff and Ord, 1973), may fail to reveal the existence of important pockets of clustering.

We have developed a parallel algorithm – P-Gi*(d) using TeraGrid. This parallel algorithm can solve larger-size problems than Gi*(d) sequential algorithms can handle while some super-linear speedups were achieved because memory requirements are significantly decreased by exploiting spatial data parallelism. Grid computing is particularly suitable to support typical exploratory scenarios using the parallel algorithm.

Project Team: Marc Armstrong, Kate Cowles, Shaowen Wang