What

Eigenvector Spatial Filtering: motivation, conceptualization, mathematics, and applications

Who

Professor Daniel A. Griffith, University of Texas at Dallas

When

4:00-6:00p.m. May 28, 2013

Where

Room 2321, IGSNRR

Host:

Prof. WANG Jinfeng

Brief to the report

Spatial autocorrelation is everywhere—it underlies art, is found in common games, allows such viewings as television, and can be the artifact of a particular zonation scheme. In its more serious form, it creates a number of challenges in science. Eigenvector spatial filtering is a powerful new methodology that addresses these scientific challenges by creating a synthetic proxy variable, a linear combination of eigenvectors extracted from a spatial connectivity matrix that ties geographic objects together in space, and then adding this proxy variate as a control variable to a model specification. This control variable, whose individual eigenvectors are distinct map patterns, identifies and isolates the stochastic spatial dependencies among georeferenced observations, thus allowing spatial analysis to proceed as if observations are independent. This presentation briefly outlines the motivation and development of this methodology, highlights some of its interesting mathematical aspects, and illustrates some of its advantages vis-à-vis popular spatial autoregressive model specifications.

From

State Key Laboratory of Resources and Environmental Information System