| This dissertation describes the application of recent advances in Bayesian statistics techniques to the challenge of managing uncertainty in geospatial data. A model for geospatial data production is developed that clarifies the importance of understanding and managing data quality throughout the geospatial data lifecycle. Traditional error theory for mapping applications is reviewed, and the limitations of the traditional error techniques are identified. Traditional mapping error theory can be extended, using recent advances in Bayesian statistics, into Bayesian Networks, Bayesian Hierarchical Models, Probabilistic Graphical Models, and Markov Chain Monte Carlo techniques. These techniques are applied to the problems of: estimating the quality of geospatial data, integrating geospatial data from different sources with different qualities, describing data quality in metadata, propagating data quality through Geographic Information Systems Models, and visualization of data quality by the user.{09}Example applications are presented for managing uncertainty in continuous variables, and for categorical data. Finally it is demonstrated that these techniques can be integrated as automated tools with current GIS software. |
