A Strategy for Quality Assurance of Land-cover/land-use Interpretation Results with Faulty or Obsolete Reference Data

Land cover mapping in most cases results in thematic maps which are in general not a perfect representation of the reality. They always contain errors due to the method used for its production or simply the difficulty that in many cases the land use is not equivalent to the land cover and semantic knowledge of an interpreter has to be used to derive the finally wanted object type. In order to assess the reliability of a particular land use map procedures for quality control and checking the geometry and thematic contents of the mapped objects have to be applied. The result of an accuracy evaluation typically provides the user with an overall accurcy of the map and the accuracy for each class (object type) in the map. Conducting a proper accuracy assessment of a product created using remotely sensed data can be time and resource consuming and respectively expensive, especially if no appropriate or up-to-date reference material is available and ground truth data has to be collected. In this paper a method is presented, which makes use of reference data (thematic maps), which have been produced in another context using class descriptions which slightly differ from the class descriptions used in the actual land use mapping and which have been generated some years before the actual mapping. It can be shown, that it is possible to use this “obsolete” data for accuracy checks, because means are implemented to find those areas in the reference data, which are out-of-date or which have been falsely assigned to the class which is inspected. The procedure allows to exclude these areas from the computation of accuracy measures. The method has been implemented in ArcGIS 9 using SQL-based algorithms.