Assessment of SVM and MLC Algorithms on Landuse/ Landcover Mapping of Riparian Forest, Using OLI Sensor (Case Study: Riparian Forest of Maroon, Behbahan)

Obtaining more accurate and updated information about the forest area is one of the basic factors in sustainable management of this area. Acquiring this information is more beneficial in terms of time and cost through classification of remote sensing data. In this paper, Landsat8 (OLI) data from Maroons Behbahan riparian forest that is located in Khoozestan province of Iran were used for mapping and better management of riparian forest. Preprocessing operation including radiometric and atmospheric correction was applied to the data. Supervised classification algorithms including maximum likelihood (MLC) and support vector machine (SVM) with seven and three classes were used for classification. In order to evaluate the capability of support vector machine, three categories of training data with 241, 141 and 41 numbers with four kernels of SVM (linear, radial basic function, sigmoid and polynomial) were used. The results indicate that mapping of Maroons riparian forest using Landsat images is possible and the best result was acquired using SVM –polynomial method by three classes with overall accuracy and kappa coefficient of (99/24) % and (0/97) respectively. Also, the findings showed that with reduction of number of classes from seven to three, the accuracy of classification is increased. By reducing the number of samples to moderate, significant difference in accuracy of classification was not observed, but by more reduction of samples, the accuracy of results is reduced.