Wearable backpack Light Detection and Ranging (LiDAR) systems have been widely used for high-resolution data collection in urban environments. However, they are often limited by the operator's mobility and time required for complete coverage of the area of interest. This paper introduces a BikePack LiDAR system, which uses a bicycle for efficient urban data acquisition. Despite its advantages, the system face challenges of intermittent Global Navigation Satellite System (GNSS) signal availability due to dense, tall buildings and other objects in urban environments. The study proposes a framework to enhance the trajectory and mapping results for the BikePack LiDAR in GNSS-challenging urban areas. The proposed framework offers an option to incorporate airborne LiDAR data to improve the absolute georeferencing accuracy of the derived point cloud, enabling the integration of the two sources – terrestrial and airborne, to produce a comprehensive 3D map of the urban environment. Following the enhancement, this study also demonstrates a learning strategy for isolating vegetation from other man-made and natural objects. Based on the well-aligned terrestrial BikePack and airborne Geiger-mode LiDAR data, a deep learning strategy is applied to the latter, with derived semantic segmentation results transferred to the BikePack point clouds through a cross-labelling process. The experimental results show that the proposed trajectory enhancement strategy can significantly improve the relative and absolute accuracy of the BikePack point cloud with the assistance of Geiger-mode airborne LiDAR data, achieving planimetric and vertical trajectory adjustments of 0.36 m and 0.27 m, respectively. Furthermore, the semantic segmentation results show that the proposed cross-labelling strategy outperforms other methods, improving overall accuracy by approximately 16%, and increasing the mean Intersection over Union (IoU) and Cohen’s Kappa score by 0.17 and 0.24, respectively.