Comparing the consistency of expert land cover knowledge

Data integration can be hindered by differences in data semantics and meaning. The problem is that different data encapsulate different conceptual views of the world. Integration approaches have been developed based on modelling expert opinion of how dataset relate, rather than statistical descriptions of data correspondence. But different experts have different opinions and this is a problem in the interpretation of remotely sensed data as much as in other areas of endeavour. In work reported here, the opinions of three experts were used to examine the semantics of land cover information derived from satellite imagery. We examined the integration of two land cover datasets of the same area at different dates where the land cover mapping classes are very different, and apparently incompatible. The approach adopted involves expert opinion of how the two land cover datasets relate under a scenario of idealised relations. The work reported here compares the performance of three different experts in three different scenarios, and evaluates their performance at identifying areas of land cover change. The results show that overall they identify the same parcels as potential change areas but different experts are more reliable at identifying change in specific landscape types.