Abstract:

Aims: Investigating the community characteristics of tree species composition and structure is of vital importance in evaluating urban vegetation growth status and its ecological function. The quantitative characteristics of different tree species in a community act as the basis for quantitative descriptions of the composition and structure of vegetation communities. Traditional vegetation surveys require substantial human and material resources and are therefore difficult to carry out on a large scale, other options—such as single remote sensing data source and corresponding method— frequently do not provide accurate information on the number of trees, crown size, and tree species simultaneously. To address these challenges, it is essential to explore the potential of unmanned aerial vehicles (UAV) with remote sensing technology that deploys multi-source datasets to accurately acquire the composition and quantitative characteristics of urban vegetation communities. 

Methods: The study area was located in Lushan, a famous tourist city with high vegetation density and a complex terrain. We employed high-resolution RGB imagery, LiDAR point cloud, and their combination to conduct individual tree detection and species classification. We then compared the performance of different remote sensing data sources and their corresponding methods of extracting quantitative characteristics of tree species in the urban vegetation community in order to assess their efficaciousness. 

Results: The results indicated: (1) The multi-source datasets yielded optimal results for individual tree segmentation and species classification. Compared to using imagery or LiDAR data separately, the F-score for multi-source data on individual tree segmentation increased by 0.116 and 0.102, respectively, and the overall accuracy of tree species classification improved by 12.1% and 23.1%. (2) The combination of multi-source data improved the accuracy when extracting the relative quantitative characteristics of tree species within communities, with errors for the relative density and relative coverage of all tree species categories being within 2.3% and 4.8%, respectively. In contrast, methods based on a single data source suffered from inaccuracies, either overestimating or underestimating the relative quantities of specific tree species. 

Conclusion: The research demonstrated that combining multi-source data improved the accuracy of extracting the composition and quantitative characteristics of urban vegetation communities by simultaneously optimizing the process of individual tree detection and that of species classification. Further, it offers theoretical support and a methodological framework for subsequent UAV remote sensing monitoring and its contribution to the structure of urban vegetation.

Key words: urban vegetation, tree species composition and quantitative characteristics, UAV remote sensing, high-resolution imagery, LiDAR