Mapping of understory infested boxwood trees using high resolution imagery

The Buxus Hyrcania stands in the Hyrcanian forests have been infested by Buxus blight pathogen in recent years. Determination of the rate, and finding the spatial distribution of infested trees, is very important for control and treatment management. In a case study, the capacity of leaf-off (winter) Pleiades high-resolution satellite imagery was examined using pixel and object-based classification algorithms to map the different infestation intensity classes in the reserved area of Anjilsi-Khiboos, a small part of the Hyrcanian forest. Ground truth of five infested classes including healthy, semi-healthy, full-infested, forest stands without boxwood understory and, the non-forest area was prepared using accurate positioning using RTK-DGPS. The images were geometrically orthorectified using DGPS ground control points and a DEM. Using suitable vegetation indices, the separability of classes on main and processed bands was evaluated using a transformed divergence index. Image classifications were carried out using appropriate algorithms and assessed using unused ground truth points. In the pixel-based classification, the best results were achieved using maximum likelihood algorithms on the original VNIR bands. In the object-based classification, the Bayes algorithm on VNIR bands had better performance among other used algorithms; however, the object-based classifiers had slightly the same results compared with pixel-based classifiers. The results showed that due to the similarity of spectral responses of healthy and semi-healthy boxwood trees and the heterogeneous distribution of boxwood blight disease in the study area, detection of the healthy trees from infested trees was slightly impossible. However, the fully infested and dead trees could be detected well from healthy boxwood trees and other stands and classes.