Geospatial data with high resolution and spatio-temporal accuracy can further support sustainable infrastructure and optimise urban services to improve the quality of life of city residents. LiDAR-based technologies are commonly used to produce 3D urban models and can include terrestrial laser scanning (TLS), mobile mapping systems (MMS), and airborne platforms such as photogrammetric drones. Point cloud datasets can be utilised for transportation planning and management, utility management, green infrastructure evaluation, and emergency response. Despite the utility of these point cloud datasets, the intrinsic incompleteness or sparsity due to the costs of surveying, the characteristics of the sensors, and environmental occlusion are significant limitations for effective precision modelling at the urban scale. Point cloud upsampling appears to be an innovative modelling gap for synthetically increasing point density, while preserving geometric accuracy. Deep learning–based networks demonstrably reduced the quantified improvements of the point cloud upsampling method. Previous studies have shown that reduced point-to-surface deviation from ~0.146 to ~0.140 (10^-2 scale; 6.11 % improvement), and improved distribution uniformity from 0.315 to 0.219 (30.55 % improvement), and frequency-selective geometry upsampling provided up to 4.4×s less point-to-point compared to PU-Net and at 4× upsampling factors These results demonstrate that advanced point cloud upsampling methods would reasonable improve the accuracy or precision of derived products such as digital terrain models (DTMs), canopy height models (CHMs), and other ecological indices that are generally sensitive to point density. This paper reviews the latest upsampling algorithms and proposes a way of thinking and structuring data science that can scale into urban monitoring processes.