Enhancing rooftop solar energy potential evaluation in high-density cities: A Deep Learning and GIS based approach

Abstract

The assessment of rooftop solar potential is vital for optimal photovoltaic (PV) system placement and renewable energy policy in dense urban areas. Complex shading from buildings and diverse rooftop obstacles have posed significant challenges to this evaluation. We propose a method that leverages Deep Learning and Geographic Information Systems (GIS) to precisely gauge solar energy potential at the city scale, accounting for shading and obstacle effects. We integrated building data with height attributes to measure inter-building shading and employed the DeepLab-v3 Convolutional Neural Network to identify rooftop obstructions from high-resolution satellite imagery. Using this hybrid framework, we calculated the available rooftop area in Shanghai, excluding the Chongming Island, and produced a detailed map of PV potential. Results show that the estimated annual potential for rooftop solar radiation in Shanghai stands at 257,204 GWh, with a predicted annual PV electricity generation of 49,753 GWh. In the study area, obstacles occupy approximately 14.9 % of the rooftop area. Neglecting the impact of rooftop obstructions and shading effects would result in a 25.6 % overestimation of the rooftop PV capacity. This work advances the precision of renewable energy development and informs sustainable urban planning strategies.