Comparison of two classification methods of maximum probability and artificial neural network of fuzzy Art map in making Range land cover maps (case study: Range land area of Doviraj area, Dehloran)

Rangelands are one of the most important renewable resources and because of their extent and economic, social and distinctive environmental impacts are of very special importance. Unfortunately, in our country, like most developing countries, rangelands have been exposed to degradation for various reasons including the non-systematic management of these resources. Remote sensing technology and satellite data are useful tools in the studies of rangeland and vegetation sciences. One of the applications of satellite data is mapping range land use. The aim of this study was to compare two methods of maximum probability and fuzzy for rangeland zonation. For this purpose, Landsat ETM+ was used; then, after final geometric and radiometric corrections, the final classification map was prepared. According to the results of accuracy of these two methods using the kappa coefficient, the artificial neural network algorithm of fuzzy Artmap with a coefficient of 0.9614 was more accurate than the maximum probability algorithm with a coefficient of 0.8058. Results of this study also indicated that the traditional algorithms of classification such as statistical methods due to their low flexibility, and parametric types such as maximum probability method because of the dependence on the Gaussian statistics model, could not provide optimal results, when the samples were not normal. In this study, ENVI 4.5, Idrisi Andes 15 and Arc GIS9.3 software were used.