Web-Based Cluster Analysis for the Time-Series Signature of Local Spatial Association

We propose a method for modeling the time-series of local spatial association in geographical phenomena and implement a Web-based statistical GIS for the time-series analysis using client-provided dataset. In order to examine the pattern of time-series and classify similar ones on a cluster basis, we employ Moran scatterplot and extend it to time-series Moran scatterplot accumulated over a certain span of time. Using the time-series Moran scatterplot, we develop similarity measures of “state sequence” and “clustering transition” for the time-series of local spatial association. If we connect

n

corresponding points of a region on the time-series Moran scatterplot, the connected line composed of

n

nodes and

n

-1 edges forms a time-series signature of local spatial association for the region. From the similarity matrix of the time-series signatures, we generate a map of the clustered classification of changing regions. These analytical functionalities of cluster analysis on the time-series of local spatial association are implemented in a Web-based GIS using XML Web Services.