Computer-Assisted Large Area Land use Classifications with Optical Remote Sensing

Abstract

An overview shall be given on operational methods and steps involved, when optical remote sensing data shall be digitally processed to result in a land use data base, which certainly forms one of the most prominent tasks of remote sensing. Questions of terminology (especially land use and land cover) will be covered, as well as data selection and acquisition, noise correction, gee-coding, classification, post-processing and map production. Obviously, only guide-lines can be given and it would be ways beyond the scope of this article to cover the whole spectrum of interesting approaches. It must be pointed out, that high quality demands call for an adequate regard of ancillary data; their use is still hampered by technical barriers as disperse storage and solely analogue availability, various geometric projections, and others. Moreover, commercial image processing software for use with remote sensing data does hardly provide any tools to imbed a-priory knowledge. Gee-scientific knowledge on vegetation patterns, crop-rotation systems in agriculture and phenological information around the time of image acquisition ('dynamic vegetation models') can significantly improve the classification results. The core task is the design of an efficient classification method, which must be carefully adapted to the class specifications. A 'brute-force' approach aiming at results in a single step with one universal classifier cannot be recommended. Examples for a more sophisticated solution are basically taken from a large-area project for the state of Saxony (Germany): A combination of default functions and additional procedures was allowing to profit from a selective choice of spectral bands, classifiers and (post-)processing steps at every node of a hierarchical classification tree. Wherever local image features were performing insufficiently, textural or form attributes have been included. The cited project was accompanied by the generation of a set of 15 land-use maps in a standardised layout. Finally, some remarks will be given concerning a potential project for a comprehensive land-use map in a less-developed area like Albania.