Separation of citrus plantations from forest cover using landsat imagery

The development of a National Forest Inventory (NFI) process is mandatory for Turkey as a country in the process of negotiations for European Union membership. Research is currently being under-taken into developing an appropriate model for a NFI suitable for Turkey's forest conditions. This study was undertaken within the framework of a cooperative project that explores the potential applications of satellite data for the development of a Turkish NFI. The main goal of the study is to determine the ability to discriminate between citrus orchards and other agricultural areas from forest cover using Landsat ETM+ data in a selected area in the Mediterranean Region of Turkey. Both pixel-based and object-based classification approaches were evaluated for their utility in large area classification (Figure 1 and 2). The Maximum Likelihood algorithm was used for the supervised classification, while the ISODATA algorithm was used for the unsupervised classification. A classification system based on a hierarchical schema with three levels using nearest neighbour and membership function classifiers was employed in the object-oriented classification. The accuracy of these was then compared with forest stand maps produced using 1:15,000 scale aerial photographs. The most accurate result was achieved using an object-oriented classification system (Table 1). This classification method produced an overall accuracy of 93 % and a corresponding K-hat of 0.91 for five land cover classes, which included: Water, Productive Forest, Non-Productive Forest, Citrus and Non-Forest Areas. Consequently, it is concluded that Landsat data can be employed for two objectives in Turkish NFI: i) the identification of "productive forests" and "non-productive forest" in order to determine sampling intensity, and ii) the pre-clarification of forest/non-forest area.