Regional Classification of Winter Wheat Using Remote Sensing Data in Southeastern Turkey

Abstract

Accurate and timely information about crop acreage estimation is important for agricultural management. In Turkey, wheat production is very important, and it is widely planted in Anatolia and in Southeastern Turkey. In this study, four different classification types were evaluated for wheat determination. As a study area, the region of Islahiye and Nurdagi counties of Gaziantep, Turkey was chosen. As satellite data, a Landsat 8 OLI image acquired on April 10, 2017 was used. The ground-truth points that were collected in surveying, and additionally field information taken from farmer registration system provided by local administrations were used as references. The application was done by classification of the satellite image using four different methods (Maximum Likelihood, Support Vector Machine, Condition-Based and Nearest Neighbor). After the results were obtained, the wheat classes obtained were transformed to vector format to overlay on the satellite image for visual analysis. The area of wheat class obtained from each method was presented and compared. The results were also evaluated by comparing with the data taken from Turkish Statistical Institute. All of the methods provided results close to the Turkish Statistical Institute records. Even the results were not significantly different from each other, wheat area determined using Support Vector Machine classification was better than others. The accuracy assessments were performed by calculating the total accuracy and KAPPA/KIA coefficient. The accuracy assessment analysis showed that the three supervised methods were better than the unsupervised one. As a future study, evaluation of these four classification methods using a multi-temporal dataset is planned.