Forest Boundary Mapping Using Multi-Temporal Satellite Images in Forested Areas of North Iran

Since separation of forest boundary using mono-temporal images is almost not possible, this study was performed in order to delineate Caspian forests boundaries using multi-temporal SPOT5 and ASTER images in forested areas in Northern Iran. The forests are composed of deciduous broad-leaved forests, afforested coniferous evergreen stands, tea and citrus orchards, as well as farms and orchards with deciduous species. SPOT5-HRG and Terra-ASTER images acquired between February and August 2002 were analyzed. Geometric correction (orthorectification) was applied using a precise digital elevation model. In addition, the images were geometrically co-registered. Following preprocessing, suitable training areas were determined and refined. Separation of above forest and non-forest classes was carried out using mono-temporal and multi-temporal classification approaches, as well as hierarchical and digital-visual hybrid approaches. In order to determine the accuracy of the resulting classifications, ground truth (forest boundary) data was prepared using field surveying with GPS. The total length of the forest boundary surveying route was 64.5 km. The results of the above classifications were compared with the ground truth and their accuracy was determined. The Kappa coefficient used for classification accuracy assessmen