Mapping Land Cover with Commercial and Freeware Image Classification Software – An Example from Bavaria, Germany

Abstract

This paper deals with mapping object-oriented land cover classification of digital remote sensing imagery. It is a new method of automated interpretation of remote sensing imagery. In its time the EU land cover project CORINE targeted the generation and dissemination of quality information about land cover in Europe. By now the existing data sets require updating and revision. Visual interpretation of the landscape is time-consuming. That is why fresh efforts are required to automate the interpretation process. To perform the necessary update of the land cover data set, it is important that image classification does not exceed a tolerable amount of time. Computer-based object-oriented classification is faster than visual interpretation of a human operator. Object-oriented classification is an automated process, interactive correction is, however, required to yield results comparable to visual image interpretation. Land cover maps created using this method present geo objects close to reality on the ground. In the research presented, the mapping of land cover has been performed based on high resolution digital orthophotos of the German HRSC sensor. The area of interest is located in Bavaria, in the south of Germany, on the northern fringes of the Alps in the Murnau-Staffelsee area. The method applied is compared in a commercial and a freeware environment: eCognition/Definiens vs. Spring. The lassifications have been evaluated for accuracy both visually and statistically. This paper presents a cutout of the land cover map produced including legend and discusses the potentials and limitations of the software systems used.