Towards realistic urban rooftop solar photovoltaic large-scale planning: integrating artificial intelligence-based roof detection, 3D geometry, and energy modelling

Urban rooftop solar photovoltaic (PV) systems represent a key strategy for advancing city-scale decarbonisation, yet their realistic potential is frequently overestimated due to simplified geometric and energy assumptions. This study presents a fully automated methodology based on artificial intelligence (AI), LiDAR-derived three-dimensional geometry, and solar energy modelling to identify large-scale urban rooftop PV installation potential and to generate optimised PV panel layouts that maximise annual electricity generation. The workflow integrates high-resolution georeferenced imagery, LiDAR point clouds, cadastral data, and open-source GIS tools within a modular Python environment, enabling the identification of technically viable rooftop areas and the optimisation of PV layouts with corresponding annual energy yield estimation. The methodology accounts for geometric constraints, minimum irradiation thresholds, slope and orientation filters, structural proxies, and maintenance corridors. Applied to the city of Valencia (Spain), the methodology is implemented over approximately 18.4 million m 2 of urban rooftop surface. The resulting layouts allow the installation of 342,785 PV modules, corresponding to 173 MW of installed capacity and an estimated annual electricity generation of 270 GWh. These results demonstrate the capacity of automated, spatially explicit approaches to support realistic urban energy planning and decision-making processes.