Comparison of different algorithms for land use mapping in dry climate using satellite images: a case study of the Central regions of Iran

Abstract

The objective of this research was to determine the best model and compare performances in terms of producing land<br />use maps from six supervised classification algorithms. As a result, different algorithms such as the minimum distance of<br />mean (MDM), Mahalanobis distance (MD), maximum likelihood (ML), artificial neural network (ANN), spectral angle<br />mapper (SAM), and support vector machine (SVM) were considered in three areas of Iran's dry climate. The selected<br />study areas for dry climates were Shahreza, Taft and Zarand in Isfahan, Yazd, and Kerman Provinces, respectively. Three<br />Landsat ETM+ images and topographical maps of 1:25,000-scale were used in the present study. In addition, training<br />samples for each land use were constructed using GPS and extensive field surveys. The training sites were divided into<br />two categories; one category was used for image classification and the other for classification accuracy assessment.<br />Results show that for the dry climate areas, Maximum Likelihood and Support Vector Machine algorithms with averages<br />of 0.9409 and 0.9315 Kappa coefficients are the best algorithms for land use mapping. The ANOVA test was performed on<br />Kappa coefficients, and the result shows that there are significant differences at the 1% level, between the different<br />algorithms for the dry climate zones. These results can be used for land use planning, as well as environmental and natural<br />resources purposes in study areas.