Modeling Map Positional Error To Infer about True Feature Location

• Activity Number: 432
• Type: Invited
• Date/Time: Wednesday, August 10, 2005 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Jarrett J. Barber*+ and Alan E. Gelfand
• Companies: Montana State University and Duke University
• Keywords: Bayesian model averaging ; Berkson model ; bivariate spatial process ; coregionalization ; measurement error

Abstract

We address the issue of positional error, or the difference between location as represented in a spatial database and the corresponding (but unobservable) true location in the presence of reference location information of higher accuracy than the map locations. Our data consists of feature locations on one or more maps as well as control points (so called ground-truthing) associated with a subset of the features. The control points are either Global Position System (GPS) measured locations or locations on our highest quality map. We assert that positional error is composed of three sources that we refer to as large-scale error, small-scale error, and measurement error. A key feature of our formulation is to model the true location as varying in a specified way around the observed location. Often, for a region, we have multiple maps of varying accuracy along with ground-truthing data. A natural strategy would be to use all the maps to learn about true locations and true features. But this suggests a context for some type of model averaging as true location means the same across all maps (models).