Mapping carbon stock and growth of individual street trees using LiDAR-camera fusion-based mobile mapping system

Abstract

Urban street trees account for a significant fraction of trees in urban areas, yet the amount and changes of their carbon stocks remain largely unexamined. This study introduces a framework utilizing a Light Detection and Ranging (LiDAR)-camera fusion-based Mobile Mapping System (MMS) to estimate carbon stocks in individual street trees regularly. This system allows repetitive and simultaneous collection of species information and structural parameters on a city-wide scale, enabling the estimation of carbon stock and its change. The framework comprises two principal components: the detection of individual street trees and the estimation of their respective carbon stocks. To detect individual street trees, we initially employed image-based deep learning model to diminish the effort needed in constructing point cloud training data and designing a universal rule applicable to complex and diverse urban streetscapes. In the carbon stock estimation phase, we used species-specific allometric equations based on species information derived from YOLOv3 and Diameter at Breast Height (DBH) measurements from trunk point cloud circle fitting. The proposed individual street tree detection method achieved an F1-score of 81.9 %, precision of 86.3 %, and recall of 78.5 % in city-scale experiments. Additionally, the Root Mean Square Error for the estimates of DBH and tree height (H) was 3.2 cm (11.4 %) and 1.8 m (18.3 %), respectively. Repeated acquisitions between two years revealed the median change of H, DBH, and carbon stock as 0.4 m yr⁻¹, 1.4 cm yr⁻¹, and 27.1 kgC yr⁻¹, respectively. Applying our method in most vehicle accessible streets in Suwon, Republic of Korea, we mapped 34,124 street trees, revealing a total carbon stock of 6.18 GgC. These results underscore the accuracy and scalability of the framework, highlighting its potential to facilitate efficient urban carbon management.