Comparison of Object and Pixel Based Classification Methods for Dynamic Land Cover Mapping with Landsat Multi-temporal Images

Rapid urbanization has negative effects on natural resources, environment and climate due to transformation of the forests, natural lands and agricultural areas into impervious surfaces. In developing countries, especially metropolitan cities have been facing continuous urbanization as a result of migration and industrialization. Therefore, there is an obvious need for regular monitoring of the urbanization related land cover changes for effective natural resource management and land use planning activities. Satellite images are commonly used geo-data sources for deriving the land cover information. Classification of the satellite images provides thematic land cover maps of broad areas with an accuracy constrained by the spectral, radiometric and spatial resolution characteristics of the image and efficiency of the classification algorithm. In the last decade, there is an increasing effort to map the dynamic land cover information by analyzing the data produced from multi-temporal satellite images. Dynamic land cover maps include change areas in addition to main land cover classes. Istanbul is one of the most important metropolitan city of Turkey and Europe with a population around 17 million. The city still faces population increase due to migration, which results with construction of new residential areas and transportation facilities. The main aim of this research is to produce the dynamic land cover map of Istanbul metropolitan city and to perform a comparative evaluation of object and pixel based classification algorithms in terms of mapping accuracy. For this purpose, multi-temporal Landsat 8 satellite images acquired between 2013 - 2017 were used. After pre-processing of the images, built-up index data was produced for each year and layer stacked built up index data was classified using the object based nearest neighbor algorithm and the pixel based support vector machine algorithm. Accuracy of the classification results were evaluated by use of point based accuracy assessment procedure.