Resolving discrepancies in land cover datasets
Monitoring changes in land-cover on the global scale has been made possible with satellites. A variety of end-users from environmental, financial and other backgrounds are interested in analysing the evolution of land-cover features over time in a particular region. The difficulty they face is discontinuities in ontologies between subsequent surveys that often render intercomparison impossible. In order to avoid having to start anew each time, the partners in the IST project entitled REV!GIS investigated methods for uniting the disparate datasets. Geographical Information System (GIS) experts were called in to establish the linkages between the spectral information provided by the satellite and the land-cover classifications. In the context of REV!GIS, the approach was put to the test with the British Land Cover Maps (LCMs) of 1990 and 2000. Geographers with the University of Leicester extracted all instances of differences in land-cover class between LCM1990 and LCM2000. In order to determine whether the change was real or due to disparities between the methods implemented in 1990 and 2000, the British scientists employed advanced processing techniques involving Euclidean distances and special filters. Field visits were also carried out and revealed that only 40% of the irregularities were due to actual change in land cover, while the remaining 60% were attributed to miscalculation of land-cover class. Despite this disheartening outcome, the REV!GIS consortium is confident that further research can resolve these discrepancies. To that end, the Look Up Tables (LUTs) created during the project are extremely valuable. The University of Leicester intends to incorporate the LUTs into a manual for LCM2000 users.
