Visualization of Persistent Spatial Structures

In mapping dynamic geographic processes, the focus has been on variables that change over time. Though largely neglected, revealing relatively stable geographic context is equally important. In fact, in pattern recognition, identifying time-invariant features is an essential task. Doing so in space-time mapping requires first a quantitative representation of geographic structure that influences the distribution of a specific geographic variable. The aspects of geographic structure that remain relatively unchanged through a specific period of time would be considered as time-invariant variables. Selective eigenvectors of the spatial weights matrix can be used to characterize the geographic structure of a given landscape. By applying linear mixed regression to space-time data, we can identify the common eigenvectors associated with the distribution of the geographic variable of interest over a time period. These eigenvectors, along with other associated summary statistics, provide rich information about the persistent spatial structure. Effective visualization of the complex and large amount of information is needed for the interpretation and communication spatial temporal patterns. This presentation will use a real world example to discuss the design and implementation of the methods for visualizing persistent spatial structure based on space-time eigenvector spatial filtering.