A method for landscape regionalization based on fuzzy membership signatures

The aim of this paper is to present a new method for regionalization that can be used in landscape analysis and planning. The approach in this study concentrates on the possibility to use membership functions developed from training sites as a mean to characterize whole regions. Instead of traditional classification routines a fuzzy method was developed. The method is described with an example from a study area located in Sweden. Pre-classified and unclassified satellite data were used in the study. The frequency distribution of three land cover types was calculated for three different training sites. The frequency distributions were transformed into membership functions. As such they describe the graded membership for every frequency pixel to all the training sites. Thereafter, were the membership images combined through a weighted linear combination. The result shows the membership for every location in the study area to one of the training sites. The images were also classified into one resulting image showing the crisp regions of the study area. An accuracy assessment of the method was performed. The result showed an overall classification accuracy of 86% compared to existing regional divisions of the study area.