Temporal changes and land cover prediction in the prefecture of Magnesia using Markov chains and cellular automata

Abstract

The study of the temporal changes of the land cover has been one of the major subjects of research during the last years. A large number of land cover change detection research studies employ remote sensing since satellite imagery and aerial photography provide objective evidence on the earth surface events.

In this work three Landsat images (06/1991, 06/1999, 08/2000) were employed to study temporal changes in Magnisia, Greece. The radiometric and geometric correction of the images followed by their supervised classification result in the production of land cover maps for the dates mentioned. The training of the classifiers was based on the use of LIFE96ENV/GR/580 and CORINE (1991 and 2000) data. Markov chains were applied to the 1991 and 1999 land cover maps in order to produce a prediction map of land cover distribution of 2000. The Markov prediction map was then compared to the land cover classification map of the 2000 Landsat image. Although the results obtained from Markov model were satisfactory the non spatial character of the methodological chain resulted in a thematic map with spatial discontinuities. In order to overcome the spatial discontinuity problem cellular automata was applied on the Markov based land cover map.