Producing rangeland vegetation types map using different algorithms of satellite image classification

To better managing of rangeland the vegetation map is one of major factors, because plant communities is planning units of rangeland management and vegetation map shows the current status of plant communities. This research was conducted to produce vegetation type's map using Landsat 8 image classification in Behbahan, Khuzestan province. Rangelands of the study region is warm semi steppe and winter grazing. Geometric correction of satellite image was performed by ground control points with an error of less than one pixel. Atmospheric correction of existing data using the dark object subtraction was done. Field visits for vegetation type's border controlling and sampling training area was conducted. Eight supervised classification algorithms included Parallelepiped (PP), Minimum Distance to mean (MD), Mahalanobis distance (MAH), Maximum Likelihood (ML), Neural Net (NN) and Support Vector Machine (SVM) was performed. The results showed that ML algorithm has the highest overall accuracy (87.5 percent) and kappa (0.867) and PP algorithm has the lowest overall accuracy (67.1 percent) and kappa (0.571). It is suggested that, along with digital methods of classification of satellite images, visual interpretation should be used to clarify the boundary of the obtained vegetation types map.