Extending Techniques From 'Standard' Error Theory To Categorical Data And Local GIS Operations

Citation: Wright, E.J., "Extending Techniques From 'Standard' Error Theory To Categorical Data And Local GIS Operations", Proceedings of the 2000 American Society of Photogrammetry and Remote Sensing (ASPRS) Conference Spring 2000, Washington, DC, May 2000.


Error Propagation and Weighted Least Squares Adjustment are techniques for predicting accurracies and adjusting measurement observations that are widely used in surveying, geodesy, and photogrammetric applications. These techniques cannot generally be applied to GIS data and operations because they are limited to continuous (and continuously differentiable) variables with Gaussian error distributions. This paper will demonstrate that these standard techniques from error theory can be considered to be special cases of more general graphical models that can be applied to GIS data and GIS operations. Use of the more general statistical model allows error propagation and adjustment with categorical data (or continuous data) in GIS models. The general model can be used with any continuous or discrete error distribution. Solution techniques for these more general graphical models have recently been developed by the Statistics and Uncertainty in Artificial Intelligence Communities. In many cases, solution techniques are efficient enough to be implemented on PC based GIS systems. The paper will present an introduction to the theory of graphical models, show the relationship to "standard" error theory techniques, and discuss some potential GIS applications.

For more information, contact IET