Abstract:

Tree aboveground biomass (AGB) is a key indicator of urban tree ecological value, and accurate AGB assessment is a crucial component of urban tree management. Unlike AGB assessments in natural areas, urban AGB mapping demands higher precision and efficiency; traditional satellite-based methods and field-based approaches may not meet these requirements. This study proposed a Highly Efficient Urban Tree AGB assessment (HE-UTAGB) framework based on Airborne Laser Scanning (ALS) data and tested it on the open-source Dublin ALS dataset over a 2 km² area. The HE-UTAGB framework consists of two main components: an unsupervised individual tree segmentation (U-ITS) module and an individual tree point-cloud-metrics-driven AGB assessment module (ITM-AGB). These modules addressed two technical challenges related to the use of ALS data for urban tree AGB assessment: the efficient and accurate segmentation of individual trees from ALS points and the establishment of robust models correlating tree point cloud metrics with AGB. The performance of these modules was compared with four deep learning architectures for tree segmentation and four traditional allometric equations, respectively. The results showed that: (i) The U-ITS proposed in this paper yielded marginally lower accuracy than its deep learning counterparts (a 5.25% reduction in recall for individual tree segmentation), but obviated the time-intensive processes of annotation and model retraining; (ii) For the AGB model in the ITM-AGB module, the random forest developed in this study exhibited the best performance (R² = 0.85) and robustness. This study further discussed the potential generalizability of the HE-UTAGB framework across different scenarios, offering a significant reference for future urban tree carbon storage assessment and providing support for the application of the framework in a wider range of cities.