Monitoring der Verstädterung im Großraum Istanbul mit den Methoden der Fernerkundung und der Versuch einer räumlich-statistischen Modellierung

Abstract

Global urbanization takes place mostly in congested areas in the threshold and developing countries. Against this background, the dissertation aims at the monitoring of urbanization by means of remote sensing and attempts to make a spatially explicit prognosis of land use patterns, using topological and topographical variables. Study area is the eastern half of the Istanbul metropolitan area, located on Kocaeli peninsula. The survey of previous studies on the dynamics of urbanization presents state-of-the-art techniques in this field of research. Special emphasis is given to whether the methods can be operationalized in terms of obtaining area-wide, up-to-date information at minimal costs or not. Previous attempts at spatial modelling of land use changes clearly show how the model complexity increases from the simpler statistical models (Markov models) via regression models towards the recent dynamical (system) models. After picturing the demographical, economical, and socio-geographical development of the Istanbul metropolitan area, and the natural conditions and present land use on Kocaeli peninsula, the methods part of the dissertation starts with the selection of an appropriate remote sensing technique on the basis of Landsat TM images. From two alternative approaches, digital change detection in a narrower sense and post-classification comparison, the latter is chosen because of a) the necessity of major radiometric corrections with most digital change detection techniques and b) the more comprehensible land use change in a post-classification comparison. The image processing can be characterized as a multi-component approach which does not rely on a parametric classifier. Research has shown that heterogeneous land cover, which is often associated with urban-industrial land use, causes many difficulties when using parametric classifiers. For the eastern part of the Istanbul metropolitan area, an increase in urban-industrial land use of about 10 % has been observed for the period from 1994 till 1998. Roughly half of that increase consists of industrial/commercial plants built according to regional plans. The multitemporal mapping is validated by using reference maps, derived from the delineation of aerial photos. The spatial-statistical modelling approach regards the spreading of urban-industrial land use from 1994 till 1998 as the explained variable. On the basis of the survey of previous spatial modelling, the following variables are regarded as the explanatory ones: land use in 1994, distance to urban-industrial land use in 1994, distance to main roads, altitude, and slope. All variables have been derived from the satellite images and a digital terrain model. The model uses a non-parametric discriminant analysis with five nearest neighbors. Its validity and prognostic performance is tested by cross validation using non-overlapping training and test sets. The multivariate, non-parametric model yields a prediction success of 86.9 %. However, a model within the abstract discriminant space that uses topological (distance) variables turns out to be a model that works within the concrete map space. Thus, the model does not meet the demand to answer the two questions of ""why?"" and ""where?"" regarding future land us The concluding discussion evaluates the results from the image processing and spatial-statistical modelling against the background of previoulsy cited literature. It is recommended that the knowledge base for the spatial-statistical model be extended to socio-economic data, as opposed to relying solely on data derived from remote sensing and cartography.