Bayesian analysis of thematic map accuracy data

The analysis of thematic accuracy of maps and classifications relies on the interpretation of the error matrix. The statistical basis of the analysis has to date been from a frequentist point of view. An alternative Bayesian approach, described here, has a number of attractive features. Perhaps most importantly, it allows us to combine prior information in our analyses. The interpretation of the Bayesian estimates and intervals is also more intuitive, and because our inferences revolve around a posterior distribution, we are able to summarise estimates even after complicated transformations. Two novel plots summarising the accuracy of a map are also introduced. The approach is illustrated by analysing accuracy data from a land use map for the Baffle catchment in Queensland, Australia. Analysis is carried out using a non-informative prior and an informative prior based on data previously collected for a neighbouring catchment. The use of the informative prior tended to increase the precision of key accuracy parameters.