A methodology for advanced change detection with fuzzy image classification

High-quality land cover mapping and land cover change analysis are important for regional and global studies. Time series of satellite imagery seem suitable for land cover mapping, by integrating processes on the Earth surface into the classification result. However, the coarse spatial resolution in contrast to small-scale land cover patches as well as the mapping of transition zones or inherently defined mixed classes necessitate fuzzy image classification. This study describes a methodology of fuzzy image classification using a decision tree approach and bagging. The presented bi-annual change detection technique builds directly on the fuzzy classification result and presents the change in land cover composition for each class or each pixel. Furthermore, the discrete map agreement is presented as a function of classification confidence. Results based on MODIS time series for South Africa and Germany illustrate the change detection approach.