The performance analyses of support vector machine classifiers for examination of the temporal change of land-use/cover in the Beyşehir Basin in Turkey (1984-2018)

This study aimed to investigate the temporal change in land-use/cover in the Beyehir-Kaakl Subbasin, which is one of the nine subbasins of the Konya Closed Basin and known as the largest closed basin in Turkey, using Remote Sensing and Geographic Information Systems techniques. For this purpose, in the study, Landsat Thematic Mapper, Enhanced Thematic Mapper, and Operational Land Imager digital satellite images obtained in the years 1984, 1990 The Support Vector Machines (SVM) method was applied as the classification method. In order to apply the SVM method, firstly, the kernel function and parameter set, giving the highest accuracy in the classification, were selected. In the study, four different kernel functions and different parameter sets were experienced as different from each other. Seventy-two different models in total were applied using different combinations of parameters. As a result of the trials of seventy-two different parameters, it was concluded that the method and algorithm giving the most accurate result with 83.81% classification accuracy and 0.7949 Kappa statistics were the polynomial function of SVMs. As a result of the classification process examined the period between 1984 and 2018 using the determined algorithm and parameters, it was detected that artificial surfaces increased by 418%, arable agricultural lands and pastures decreased by 14%, forests and semi-natural areas increased by 4%, and coastal wetlands on the coasts increased by 6%. On the other hand, the surface area of the water bodies in the region, which demonstrated a decreasing trend until the year 2003, was determined to increase by 3% with the establishment of Sula Storage in 2003.