Urban vegetation, particularly trees, plays important roles in the urban ecosystems. In this study, we examined the potential of WorldView-2 imagery (acquired on September 14, 2012) for urban tree species classification in the capital city of Beijing, China. Four tree species including Chinese white poplar(Populus tomentosa Carrière),Chineses scholartree(Sophora Japonica), Gingko(Ginkgo biloba L.)and Paulownia(Paulownia Sieb.) were identified. To evaluate the impact of complex urban environment on urban tree species classification, we compared classification accuracies on shadow-removed image and shadow-recovered image. Object-based hierarchical approach was used to detect shadow/non-shadow area and vegetated/non-vegetated area.
Support-Vector-Machine method was used to classify tree species in vegetated area using object-based approach. We used Linear Correlation Correction(LCC) method to restore spectral information under shadowed area. The results show that tree species classification for shadow-removed imagery obtained overall accuracy of 80.97% and kappa value of 0.7286, while the accuracy and kappa value decreased to 76.88% and 0.6693when shadow-recovered image was used. In future research, we will explore the method to improve the accuracy of classification for shadow-recovered image.