André Skupin* - San Diego State University
The tri-space of geographic features, attributes, and time has for a number of years received attention within the geographic data mining community as a basis for database creation and spatio-temporal analysis. This paper puts forth a proposal to make a more general tri-space concept the basis for systematic visual exploration of multitemporal and multivariate data. Locus, attribute, and time (LAT) are the basic elements in this tri-space. A locus could refer to the identity of a particular geographic entity, but such diverse entities as persons or documents could also serve as loci. Accordingly, attributes are as diverse and range from census attributes to linguistic expressions. When the various attributes for a set of loci are observed over multiple time periods, an LAT tri-space con be constructed. Interestingly, that tri-space can be visually explored in many different ways. What emerges in the process provides not only means to categorize existing multivariate visualization techniques but more importantly points towards new ways to systematically explore the high-dimensional tri-space.
Implemented for a 40-year data set of crime statistics for the United States it is demonstrated that the tri-space itself, without using any additional information such as geographic location, supports the construction of six configurations based on either tri-space corners or edges. Combined with dimensionality reduction this leads to six very different, yet congruent and equally holistic perspectives on the very same data set.