Identification of Tree Species by Object-Based Classification of Digital Aerial Images

Remote sensing and image interpretation have been utilized in forestry management for many years and offer great potential for vegetation mapping, especially with the availability of higher-resolution imagery. In recent years, digital cameras have expanded in their utility as an efficient tool for forest inventory and mapping. This study is a contribution to assess the high-resolution digital aerial imagery for semi-automatic analysis of tree species identification. To maximize the benefit of such data, the object-based classification was conducted in two even-aged mixed forest plantations. The first study area, dominantly consisting of chestnut-leaved oak, loblolly pine, with blackberry shrubs and occasionally distributed Arizona cypress trees, and the second study area is more heterogeneous in species composition, stocking density and canopy structure and dominantly consisting of Caspian locust, velvet maple, white mulberry, common alder and smooth-leaf elm. Two subsets of UltraCamD images were geometrically corrected using the aero-triangulation method. Some appropriate transformations were performed and utilized. Segmentation was conducted stepwise at two levels and a hierarchical image object network was constructed. The classification hierarchy was developed and the Nearest Neighbor classifier, using an integration of different features was performed. Training samples and ground truth maps were prepared through fieldwork. Accuracy assessment of the resulting maps in comparison with reference data showed overall accuracies and Kappa Index of Agreement of 90.2%, 0.82 (Area1) and 69.8%, 0.49 (Area2), respectively. Transformed images were advantageous to improve the results. The lower accuracy in Area 2 can be attributed to a high diversity and heterogeneous mixture of species. Because of the limitations of using only optical data and the potential future role of LiDAR integration, more detailed and accurate mapping of tree species would be fulfilled by applying precise 3D data, which are derived from LiDAR. The accuracy of detailed vegetation classification with very high-resolution imagery is highly dependent on the segmentation quality, sample size, sampling quality, classification framework and ground vegetation distribution and mixture.