Authors: André Skupin*, San Diego State University, Timothy Schempp, San Diego State University, Kyle Jones, San Diego State University
Topics: Geographic Information Science and Systems, Spatial Analysis & Modeling, Cartography
Keywords: geographic information science, multivariate data, spatio-temporal analysis, visualization, machine learning, web applications, crime, fire, land cover, agriculture
Session Type: Paper
Presentation File: No File Uploaded
Multivariate and multitemporal data are bountiful in contemporary geography. Examples include hyperspectral land cover monitoring, longitudinal medical studies, climate modeling, and multi-year crime statistics. Geographic information scientists have put forth a number of principal approaches to formally represent such data with a view towards pattern discovery and decision support. These have include the “triad,” “pyramid” and “three-domain” frameworks. Meanwhile, the “tri-space” approach represents an effort at systematic reconfiguration of multivariate observations for discrete entities at multiple times. In the tri-space, an individual observation exists at the intersection of three arrays, typically corresponding to three elements: locus, time, attribute. Unlike in other frameworks, a locus does not necessarily correspond to a geographic location or object, but merely denotes an entity that has a distinct identity in some physical or abstract space. The tri-space then proceeds to define objects whose identity is defined by a particular combination of one or two of these arrays, with the remaining arrays defining an object’s location in a high-dimensional space. In this manner, six different perspectives can be constructed, to which common multivariate analysis and visualization techniques can be applied. We discuss recent advances of this approach, with particular focus on machine learning and visualization workflows for monitoring complex dynamics in agricultural production, post-wildfire recovery, and crime.
To access contact information login