Assessment of a Semantic Statistical Approach to Detecting Land Cover Change Using Inconsistent Data Sets

Abstract
A semantic, statistical approach to reconciling data with different ontologies is introduced. It was applied to UK land cover datasets from 1990 and 2000 in order to identify land cover change. The approach combined expression of expert opinion about how the semantics of the two datasets relate with spectral homogeneity metadata. A sample of the changes identified was assessed by field validation. Change was identified in 41 percent of the visited parcels, and all of the false positives were found to be due to classification error in either dataset. Thus, the approach reliably identifies inconsistency between two datasets, and the results indicate the suitability of uncertainty formalisms. The inclusion of extensive objectlevel metadata by the data producers greatly facilitates practical solutions to problems of data interoperability.