Mapping individual tree crowns to extract morphological attributes in urban areas using unmanned aerial vehicle-based LiDAR and RGB data

Mapping individual tree crowns (ITCs) along with their morphological attributes provides foundational variables for estimating functions, such as thermal stress and carbon emissions, within the urban environment. However, to calculate morphological attributes, it is necessary to delineate ITCs, and applying the watershed segmentation (WS) algorithm, commonly used in forests, to urban environments presents challenges due to the reliance on single-band data and complexity of heterogeneous urban elements. Additionally, deep learning (DL) models, which excel in image analysis, are constrained by the labor-intensive label generation process. This study introduces a novel framework integrating machine learning (ML)- and DL-based approaches to map ITCs and extract the morphological attributes in urban areas using unmanned aerial vehicles (UAVs). Using ML-based approaches, we conducted object-based image analysis to optimize the use of the WS algorithm because applying WS directly to urban environments, as in forests, overestimated the ITC size by 151.37 %. This approach also improved the effectiveness of the DL model, Mask R-CNN, by resolving challenges in label generation. Mask R-CNN delineated ITCs with an accuracy of 0.942, suggesting its robustness in handling the heterogeneity of tree arrangements. The results demonstrate that the proposed framework is applicable for use in urban areas globally that have similar ecological conditions. ITCs with morphological attributes provide fundamental variables for evaluating ecological functions, which are scalable for improving urban environmental planning. However, UAV data may face time and cost limitations as the monitoring coverage expands, which should be considered when applying this framework.