Comparison of visual and digital interpretation methods of land use/cover mapping in Ardabil province

Land use/cover mapping is one of the most common applications of remote sensing data. Remote sensing data by providing updated digital information, repetitive coverage, reduce costs and the possibility of processing and high potential for the preparation of land use/cover maps in natural resources, is of paramount importance. In this study, the land use and cover map prepared using Google Earth and the Operational Land Imager image sensor (OLI) of Landsat 8 satellite and methods of visual interpretation (GE images), supervised classification, neural networks and object-based classification methods (Landsat 8 images), and compared with each other. In order to evaluate the accuracy of the classification, the overall accuracy, Kappa coefficient, producer’s accuracy and user’s accuracy were used. The results showed that the visual interpretation method with overall accuracy and Kappa coefficient of 99.4 and 0.99, in comparison to the object-based, supervised and artificial neural networks (with an overall accuracy of 94, 82 and 60.8, and a Kappa coefficient of 0.92, 0.77 and 0.50) are more reliable. According to the map of visual interpretation, the rangelands with an area of 946687 ha and water bodies in the area of 217.42 ha were the largest and smallest land use/covers, respectively. In terms of accuracy, the visual interpretation method using Google Earth images had the highest accuracy, but it is time-consuming and cost-effective. In contrast, object-based method with acceptable accuracy and with low cost and time is the best method for land use/cover mapping.