A novel 3D-geographic information system and deep learning integrated approach for high-accuracy building rooftop solar energy potential characterization of high-density cities

Publication Name

Applied Energy


Accurate rooftop solar energy potential characterization is critically important for promoting the wide penetration of renewable energy in high-density cities. However, it has been a long-standing challenge due to the complex building shading effects and diversified rooftop availabilities. To overcome the challenge, this study proposed a novel 3D-geographic information system (GIS) and deep learning integrated approach, in which a 3D-GIS-based solar irradiance analyzer was developed to predict dynamic rooftop solar irradiance by taking shading effects of surrounding buildings into account. A deep learning framework was developed to identify the rooftop availabilities. Experimental validations have shown their high accuracies. As a case study, a real urban region of Hong Kong was used. The results showed that the annual solar energy potential of the entire building group was reduced by 35.7% due to the shading effect and the reduced rooftop availability. The reductions of individual buildings varied from 13.4% to 74.5%. In spite of the substantial reductions of the annual solar energy, the shading effect could only slightly reduce the peak solar power. In fact, the annual solar energy reduction could be five times higher than the peak solar power reduction. Further analysis showed that simple addition of the respective solar energy potential reductions, caused by the shading effect and the rooftop availability, tends to highly overestimate the total reduction by up to 26%. For this reason, their impacts cannot be considered separately but as joint effects. The integrated approach provides a viable means to accurately characterize rooftop solar energy potential in urban regions, which can help facilitate solar energy applications in high-density cities.

Open Access Status

This publication is not available as open access



Article Number




Link to publisher version (DOI)