Comparing the accuracy of time series classification of Landsat images in monitoring land use change

In this research, artificial neural network, maximum likelihood and minimum distance classification methods for analysis of land use changes, during 1989 to 2015, were evaluated and compared images from three Landsat satellite sensors in Sari. After geometric and atmospheric corrections, images of 1989, 2002, and 2015 were categorized under three artificial neural network algorithms, maximum likelihood and minimum distance in five land use classes. After assessing the accuracy of the methods, the Kappa coefficients were calculated for maximum likelihood, artificial neural network and minimum distance of 1989 were 92%, 87% and 65% in 2002, were 89%, 87% and 60%, and in 2015 were 91% %, 90% and 73%, respectively. These coefficients indicate the superiority of the maximum likelihood method in comparison with the other two methods in 1989. Also, the results of land use change over the whole period of the survey (from 1989 to 2015), showed that the areas of residential and irrigated lands were increased by 3615 and 575 hectares, but bare lands, gardens and forests were decreased to 1791, 1127 and 1272 hectares, respectively. According to the results, the two methods of maximum likelihood and neural network were more suitable for land use classification. The maximum likelihood method was better than the neural network method with a difference of 5% in 1989 and 2% in 2002 and 1% in 2015 in the Kappa coefficient.