Urban Tree Species Classification Using Multitemporal Airborne Hysperspectral Imagery

Tree species classification is an essential task for ecologists and forest specialists because it helps to accurately monitor biodiversity fluctuations. It becomes especially critical with the advent of highrates of biodiversity loss. The ability to monitor and evaluate the distribution of various species is an important step to ensure the stability of natural biodiversity. For ecologists, the task of creatingdistribution maps of different tree species involved sending groups of people out in the field to collect location and metadata information about those trees. However, this process can be costly andtime-consuming. In this study, we focused on classification of tree species in urban forests since these are essential parts of the urban ecosystem. With technology advancement in spectroscopy andinformation extraction algorithms, we applied machine learning approach on airborne hyperspectral images to build a high-accuracy classification model that can be used to map urban tree speciesdistribution. The classification model was trained and tested on images collected in August, September, and October 2018 with a total of over 300 000 pixels representing 11 different tree species.Our study revealed the potential of using airborne data for tree species classification as our models achieved overall accuracy of 90-93