comparison of different support vector machine kernels for monitoring of the last decade of the iranian part of eastern caspian lake

With the drop in the water level of the caspian lake in recent years, some part of this water body have become dry lands. the purpose of this research is to provide a method with a low error rate and a reasonable computational cost in order to prepare a map of changes by remote sensing techniques. in this research landsat-8 satellite data were used to extract water areas, under the effect of water indices and with the help of support vector machine. first, the images of the eastern region of the caspian sea, which is located in iran, were prepared, then classified images of two classes of water and land were prepared with the help of the desired bands and water indices with support vector machine processing. various kernels and indices were used to achieve the most accurate classification method. the classified images were compared with ground truth data to evaluate the classification accuracy. the overall accuracy of the classification in the basic mode with the wri index and linear kernel was 95.88% and in the best method with the help of the awei index and the radial basis function kernel was 97.61%. in the results of the evaluation of the accuracy of change monitoring, the optimal performance of this method was 96.21%. finally, the optimum classification approach was selected. for future researches, it is better to use this method to examine smaller scale areas.