Capability investigation on spectral images of Ikonos from leaveless season for Box (Buxus hyrcana Pojark.) understory distribution mapping in the Hyrcanian forest (Case study: Khiboos-Anjilsi Buxus reserved area, Mazandaran)

As one of the most important understory evergreen species in Hyrcanian forests of Iran, information on the distribution of Box (Buxus Hyrcana Pojark.) are essential for both forest research and practice. Here, the capability of very high spatial resolution IKONOS satellite imagery acquired in leaf-off condition was tested for mapping Box distribution in a part of Khiboos-Anjili forest reserve in Mazandaran province. The IKONOS imagery was geometrically corrected with a georefrenced panchromatic Pleaides scene, which was orthorectified using 3D ground control points obtained using differential GPS (RMSE less than one pixel). Reference data samples from three classes of non-forested area, deciduous stands without Box understory and deciduous stands with Box understory were recorded using DGPS-supported field survey. By means of a number of vegetation indices, classes seperabilities were evaluated on main and synthetic image channels by partitioning 75% training area and transformed divergence. IKONOS image was classified using both main and best-selected bands and a number of nonparametric (Maximum Likelihood, Mahalonobis distance, Minimum distance to mean and Paralell piped) and parametric (Suport Vector Machine) classifiers. Then the classified images were assessed using 25 percent of unused sample points. Results of validation using the 25% left-out test data showed the highest performance by SVM algorithm compared to other algorithms, with overall accuracy and Kappa coefficient of 97.87% and 0.96, respectively. The results also showed the potential of IKONOS imagery from leaf-off season has to map Box trees in understory layer.