| The accuracy of spatial databases has been a major focus of recent research within the GIS community. Different methods have been proposed and used to simulate errors in spatial databases. However, major differences exist as to the sources and structure of the spatial dependence of error, and concomitantly about how to model it. These differences appear to arise from conceptual differences about the nature of the error term, and the relationship between the spatial autocorrelation in the variability of the measured random variable (such as topography) and the spatial autocorrelation in error.There are many sources of error in spatial data. These sources of error have unique characteristics, requiring different approaches and methods to assess, evaluate, and model the error. Without a clearly defined and agreed upon typology of error it is difficult to communicate and coordinate among researchers, developers, and users of GIS.Several approaches are currently used to address spatially autocorrelated error, including simultaneous autoregressive models, swapping techniques, conditional simulation, and spatial filtering. However, they have not been incorporated into routine GIS operations, generally because they are too cumbersome--either difficult for most analysts to understand or too computationally burdensome. Initial investigation indicates that the underlying analytical techniques used in conditional simulation and spatial filtering offer a good starting point for the development of methods for modeling spatial autocorrelated error.This dissertation contributes to current research in GIS in three ways. First, it attempts to clarify the understanding of autocorrelation in a variable and in its error term. Second, it evaluates the advantages and limitations of various methods used to simulate spatially autocorrelated error. It also reveals the common mathematical logic and procedures used in spatial filtering and conditional simulation. Finally, a major methodological contribution is to provide users with an easily understood, well documented, and efficient procedure for evaluating and modeling unconditional, spatially dependent error. |
