Unmixing polygons in resource mapping: a Bayesian integration of expert opinion and data
by
Robert Denham
The University of Newcastle
Coauthors: Kerrie Mengersen

Mapping vegetation communities and other natural resource problems often leads to the delineation of polygons which consist of a mixture of different classes. Consequently, the resulting product becomes difficult to use as an input to further modelling work, and is often simplified to a single class, resulting in the loss of information. Statistical modelling of the vegetation classes would appear to be preferable approach, but the complexity of the product and limitations of field data mean this is rarely viable. The complexity can be overcome by using the existing vegetation maps and constructing models on a polygon by polygon basis, rather than modelling the region in entirety. The lack of field data can be addressed by using expert opinion and Bayesian modelling, which provides a formalised approach to the incorporation of expert knowledge with field data. This paper outlines such a procedure to ``unmix'' a vegetation map of Northern Queensland, and shows how this approach results in a product which provides more information than the original map with the benefit of a probability statement attached to each classification. The key limitation of this procedure is the assumption that the initial polygon boundaries are accurate and meaningful. We discuss improvements that will decrease the reliance on this assumption.

Date received: November 4, 2001

Copyright © 2001 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # caic-24.