Comparison of the accuracy of pixel-based and object-oriented methods in land use classification (case study: Samalghan Watershed)

Planning and optimal use of resources and controlling and unprincipled changes in the future, requires studying the extent of change and destruction of resources. In fact, planners for principled decisions must have a full knowledge of land use, detection, prediction of land use change and land cover in order to better manage natural resources in the long time. The aim of this study was to evaluate the accuracy of different supervised classification algorithms of basic and object-oriented pixels in land use extraction in Samalghan watershed in three periods of time 1987, 2002 and 2019. The results showed that the support vector machine algorithms for the images of 1987 and 2019 and the neural network for the 2002 image in the pixel-based classification method have the highest overall accuracy and kappa coefficient. Also, the most obvious change that can be seen by comparing the prepared user maps is the change in the level of land uses with the growth of residential areas, thus this expansion has been continuously associated with a decrease in rangeland land use. Thus, from the years of 1987 to 2019, the residential land use area increased by more than 9.197 km2 and dryland lands during these years increased by 130.89 km2, irrigated agricultural lands also increased from 44.45 km2 and Rangeland use has also decreased by 272.3 km2.