A multi-temporal classification approach to monitor a countertrend of urbanization in an era of economic crisis

Περίληψη

In the majority of rapidly developing economies, significant land use changes are being witnessed. In such cases there is an increase in the rate of the migration from the countryside to the city, while higher income groups move to newer residential areas leaving the center of the cities to be populated by lower status residents and the newest immigrants. The newer residential areas are situated farther away from the center of the cities causing an expansion of the suburban areas. Urbanization and suburbanization imply the physical growth of a city and the expansion of the built-up suburban areas. In order to monitor the urban extent constantly, the mapping community has resolved into using Remote Sensing data which covers the surface of the Earth on a regular base. Very high spatial resolution imagery (e.g. Ikonos, Quickbird, Worlview) has made possible the track of changes on a higher level of detail that is necessary to urban planners. However classification of urban areas still remains challenging due to the high heterogeneity of materials and urban objects in terms of size, shape and texture. In the present work we propose a multi- temporal classification approach for monitoring and mapping urbanization based on Hidden Markov Models (HMMs). Multitemporal satellite images enable the monitoring of the rate of urbanization by inferring thematic information from frequent image acquisitions. Our approach exploits the spectral and temporal information content on high resolution satellite data in order to minimize classification errors. It uses HMMs for representing the evolution of urban sprawl and determines the most probable sequence of classes over time by estimating class transition possibilities. The performance of our approach is tested by using a set of six very high resolution images over a suburban area of Greece covering a period of 13 years (2001-2013).