Identification of the area under cultivation of Saffron using Landsat-8 temporal satellite images (Case study: Torbat Heydarieh)

The aim of this research is the identification of Saffron fields with Landsat-8 Satellite images in Torbat Heydarieh. In this regard, two approaches were utilized. The first approach was dealing with implementation of neural network, support vector machine, Mahalanobis distance, the minimum distance, maximum likelihood and parallelepiped classification methods to achieve land cover map. The second approach was to use normalized difference vegetation index (NDVI) and soil adjusted vegetation index (SAVI) in the greenness peak time range of saffron. To prepare field data, coordinate and land cover class of 2587 points (1463 as training sample and others as tested) in a region with at least 30 m same land cover on January 25th, 2015 and May 9th, 2015 were recorded using a GPS receiver. Furthermore, statistics presented by ministry of Agriculture Jihad in the 2014-2015 crop year was used for evaluation. Two measures, including Kappa coefficient and overall accuracy were used for evaluation of the results. Support vector machine classification with overall accuracy of 95% and a Kappa coefficient of 90%, was the best method of the first approach. It shows a difference of about 18% in saffron area comparing with Jihad statistics. On the other hand, NDVI as the best method of second approach shows an area of 7118 hectares which comparing with Jihad statistics (7550 hectares), shows the error of 5.7%. Hence, the results indicate the performance of temporal vegetation indices in identification of saffron fields according to its phenology.